Increasing mass spectrometry throughput using time-encoded sample multiplexing

Increasing mass spectrometry throughput using time-encoded sample multiplexing | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Increasing mass spectrometry throughput using time-encoded sample multiplexing View ORCID Profile Jason Derks , View ORCID Profile Kevin McDonnell , View ORCID Profile Nathan Wamsley , Peyton Stewart , View ORCID Profile Maddy Yeh , View ORCID Profile Harrison Specht , View ORCID Profile Nikolai Slavov doi: https://doi.org/10.1101/2025.05.22.655515 Jason Derks 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jason Derks For correspondence: jderks{at}parallelsq.org nslavov{at}parallelsq.org Kevin McDonnell 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Kevin McDonnell Nathan Wamsley 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nathan Wamsley Peyton Stewart 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Maddy Yeh 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Maddy Yeh Harrison Specht 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Harrison Specht Nikolai Slavov 1 Parallel Squared Technology Institute , Watertown, MA 02472, USA 2 Departments of Bioengineering, Biology, Chemistry and Chemical Biology, Single Cell Center, Northeastern University , Boston, MA 02115, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nikolai Slavov For correspondence: jderks{at}parallelsq.org nslavov{at}parallelsq.org Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Liquid chromatography-mass spectrometry (LC-MS) can enable precise and accurate quantification of analytes at high-sensitivity, but the rate at which samples can be analyzed remains limiting. Throughput can be increased by multiplexing samples in the mass domain with plexDIA, yet multiplexing along one dimension will only linearly scale throughput with plex. To enable combinatorial-scaling of proteomics throughput, we developed a complementary multiplexing strategy in the time domain, termed ‘timePlex’. timePlex staggers and overlaps the separation periods of individual samples. This strategy is orthogonal to isotopic multiplexing, which enables combinatorial multiplexing in mass and time domains when paired together, and thus multiplicatively increased throughput. We demonstrate this with 3-timePlex and 3-plexDIA, enabling the multiplexing of 9 samples per LC-MS run, and 3-timePlex and 9-plexDIA exceeding 500 samples / day with a combinatorial 27-plex. Crucially, timePlex supports sensitive analyses, including of single cells. These results establish timePlex as a methodology for label-free multiplexing and combinatorial scaling of the throughput of LC-MS proteomics. We project this combined approach will eventually enable an increase in throughput exceeding 1,000 samples / day. Download figure Open in new tab Introduction Liquid chromatography-mass spectrometry (LC-MS) proteomics methods can achieve accurate quantification while analyzing over 50 samples per day 1 – 3 . This has been made possible due to advances in liquid chromatography that have enabled robust and repeatable separations of chemical analytes from complex mixtures 4 – 7 , mass spectrometry instrumentation and data-acquisition 8 – 13 , and data interpretation approaches 14 – 17 . Together, these advances have positioned LC-MS proteomics as a powerful tool in quantitative biology. However, the scale at which such datasets can be generated remains constrained by time and costs associated with LC-MS data-acquisition. As such, there is need for increased throughput of data acquisition to generate datasets that will empower biological investigations. Strategies to increase the throughput of LC-MS proteomics generally aim to 1) increase the number of LC-MS runs completed per unit time 2 , 3 , 18 , 19 , or 2) increase the number of samples acquired per run as achieved through sample-multiplexing 20 , 21 . Both strategies are complementary approaches to increase LC-MS throughput 22 – 24 . In particular, this work will focus on a new approach for the latter – sample-multiplexing to increase LC-MS throughput. Sample-multiplexing is commonly achieved through chemically barcoding samples with isotopic mass offsets to be pooled and acquired in parallel within the same run. The samples are separable at MS1-level with isotopologous mass offsets such as mTRAQ, PSMTag, or SILAC among others 25 – 30 , or at the MS2- and MS3-level with isobaric mass-tags such as TMT and iTRAQ 31 , 32 . While 35-plex TMT currently offers the highest multiplexing capacity 33 , quantifiable proteomic coverage per unit time is fundamentally limited as ‘reporter ions’ cannot be used to quantify precursors of chimeric spectra. Thus, quantification based on reporter ions is limited to quantifying one precursor per MS2 scan. Alternative approaches such as data-independent acquisition (DIA) 34 , 35 parallelize precursor analysis through the intentional generation of chimeric spectra, enabling greater depth of coverage per unit time; such approaches are amenable to non-isobaric isotopologous multiplexing 36 . Indeed, multiplexed DIA (plexDIA) has been shown to multiplicatively increase throughput at high depth of coverage and accuracy 20 , 21 . Sample multiplexing can be further increased based on a complementary and relatively unexplored dimension for multiplexing proteomics samples: time. Samples can undergo parallelized and independent separations which are offset in time, encoding sample-specificity in the time-domain. Temporal encoding has been implemented for sample multiplexing based on multisegment capillary electrophoresis (CE) and targeted analysis of relatively few analytes 37 – 39 . Here, we explore the potential of using time-domain multiplexing for parallelized data acquisition using DIA. This requires new approaches to interpreting the resulting multiplexed mass spectra. Such an approach may scale not only sample throughout, but also throughput defined as quantified proteins per unit time. Here we demonstrate two implementations of timePlex that enable the acquisition of three samples in parallel (3-timePlex). To interpret the unique mass spectra generated by timePlex, we developed a software module in JMod 40 . Furthermore, we applied combinatorial multiplexing of 3-plexDIA and 3-timePlex to acquire 9 samples in parallel per LC-MS run, and 9-plexDIA 30 , 40 and 3-timePlex for 27-plex analysis. The two implementations establish a framework for multiplexing samples in time that we expect will expand to higher plex, generalize to other implementations, and enable a combinatorial increase in proteomics throughput when paired with isotopic labeling approaches such as plexDIA. Results Multiplexing in time and mass domains The timePlex concept and its relationship to other LC-MS approaches is illustrated in Fig. 1a with elution curves for three precursors. Acquiring data by non-multiplexed DIA limits the rate of data acquisition to a single sample per run. Multiplexing in the mass dimension through mass-encoded offsets, as performed in plexDIA, enables a linear increase in the rate of sample acquisition with plex. Here we introduce a complementary approach, timePlex, by multiplexing samples using time-encoded offsets to likewise increase sample-throughput. Due to their orthogonality, timePlex and plexDIA can be used in combination to increase proteomics throughput multiplicatively. Download figure Open in new tab Figure 1 Multiplexing in the time and mass domains is combinatorial a Illustration of precursor elution curves for no multiplexing, plexDIA, timePlex, and combined timePlex & plexDIA data acquisitions. Combinatorial timePlex & plexDIA enables multiplicative scaling of the number of samples acquired / run. b Estimated cost for LC-MS/MS analysis per sample for different levels of multiplexing. The total cost per sample is estimated assuming $150/hr for LC-MS/MS, approximately $1k/column with an average 1,000-run lifespan, and the cost to label 1 µ g of protein with commercially available mTRAQ (approximately $0.36). Multiplexing nearly proportionally reduces the cost of data acquisition because the expense of LC-MS instrumentation and operation exceeds that of mass tags and LC columns. timePlex and plexDIA may be used in combination, increasing throughput and cost-saving multiplicatively with plex, as shown in Fig. 1b . Here, we seek to develop an approach for timePlex proteomics and benchmark its combinatorial potential when paired with plexDIA. Implementing timePlex proteomics data acquisition Multiplexing through temporal encoding can be implemented through several distinct approaches. Here, we aim for a proof of principle with two accessible implementations with different column types illustrated in Fig. 2a and expect significant improvements in future implementations. In both approaches the gradient flow originating from a single LC-system is split to separate columns. Time offsets are encoded by adjusting the volume of each sample’s respective capillary transfer line. In the first implementation, termed ‘single-emitter’, the flow from each column is rejoined to a single emitter for ESI. In the second approach, termed ‘separate-emitters’, the flow remains separate and they undergo separate ESI from their respective emitters. In both cases, the resulting MS data will include multiple observations of the same peptides, separated by systematic time offsets which are encoded by the different travel paths, Fig. 2b . Download figure Open in new tab Figure 2 Enabling timePlex analysis through accurate RT prediction a Design showing two implementations of timePlex. The time-offsets are encoded through differences in the volumes of the transfer lines. b Conceptual illustration of repeated observations of peptides species systematically offset in time. c Illustration of RT predicting model. Peptide sequences were one-hot encoded and a hybrid convolutional neural network and deep neural network architecture was trained to predict retention times for label-free (LF) and mTRAQ-labeled peptides. Models were trained on 91,657 LF peptides or 56,645 mTRAQ-labeled peptides. d Test data, which were withheld from training, were used to evaluate the performance of the RT predictor. e To evaluate the model’s ability to fine-tune RT’s from new data, residuals were computed between the original RT predictions of the model (shown in blue), and after finetuning on approximately 3,200 peptides (shown in orange); the plotted data is the test-set withheld from fine-tuning. The 90th percentile boundaries are shown with orange and blue lines. f The original library iRTs predicted during DIA-NN in-silico library generation of LF data were used to visualize a typical starting point of RT accuracies, as shown in the left panel. For each run, our model is updated on the initial search results to fine-tune RTs to be used in the final search. The final search output is shown in the right panel with the library RTs for the middle timePlex channel. A precursor-set ([TRPVVAAGAVGLAQR]+3) is highlighted in pink to show it’s library RT before (left panel) and after (right panel) transfer learning. In these implementations, samples are sequentially loaded and travel through the same sample-loop and parts of the same transfer lines to reach their designated columns during sample-loading. Schematics for implementing a 3-timePlex are shown in Extended Data Fig. 1 . During sample loading, peptides may bind unspecifically to surfaces in the common flowpath, leading to sample-carryover into unintended columns. To mitigate or prevent this all together, samples were resuspended in 0.015% DDM prior to sample-loading. Carryover was undetectable for the vast majority of peptides; when detected for a few peptides, it was limited to a median of 5.7% of the intensity of the intended time-channel, Extended Data Fig. 2 . Interpreting timePlex proteomics data Interpretation of timePlex data depends on highly predictable elution times of analytes (e.g., peptides). While retention time (RT) offsets are less precise and reproducible than m/z offsets, they can afford reliable sample assignment if they are engineered to be large enough to account for the accuracy of RT prediction. To analyze time-encoded multiplexed data, we developed a specialized timePlex module in JMod 40 . The module implements (1) time-channel detection and (2) main search. The time-channel detection determines the times when precursors from each channel are expected. Then, these times are used to perform the main search. The time-channel detection starts by identifying peptide-like features that correspond to the same peptide sequence detected in each of the n samples. The temporal order of the n observations are then used to build retention time expectations for each time-channel. These initial identifications are then used to finetune the retention-times of the initially provided library, the results for which are then aligned to fit each time-channel, which is then used for the main search. This approach remains flexible to differences in time-offsets between multiple time-channels, as this process is automated. As a result, the user does not need to specify the time spacing between samples, only the number of samples multiplexed in time to expect. While precursors across time-channels are systematically separated in time, accurately assigning each identified precursor to a sample becomes challenging when a given peptide is not observed in all time-channels. This difficulty stems from inaccuracies in library retention times, which we improved by creating a retention-time predictor as others have done 41 – 48 to be used for fine-tuning library retention times on a per run basis 49 . The model architecture implemented in this work is a convolutional neural network (CNN) ending in fully connected layers of a feedforward deep neural network (DNN) as illustrated in Fig. 2c . A summary of the model architecture is shown in Extended Data Fig. 3 . LF and mTRAQ models were trained on fewer than 100k peptides each; test-data was withheld to evaluate the accuracy of the models for unseen mTRAQ and LF peptides. We observed strong agreement between predicted and empirical iRTs ( R 2 > 0.997) for both labelfree and mTRAQ-labeled peptides, as shown in Fig. 2d , which suggests these models accurately predict retention times, and may be suitable for transfer learning. Data generated outside of the training and test sets was used to evaluate the potential for these models to refine library retention times given new data acquired under different conditions (different gradient, instruments, and column batch). We found that the original model was able to predict 90% of the data within 30.0 seconds of the empirical RT, and after fine-tuning it improved to within 8.8 seconds as shown in Fig. 2e . The improved RT accuracy stands to benefit DIA search performance due to increased specificity of retention time search spaces, even in non-timePlex applications, as previously shown by Wallmann et al. 49 Fine-tuning retention times on a per run basis is of even greater importance with timePlex data where inaccurate library retention times will lead to ambiguous time channel assignment of peptides, as shown in the left panel of Fig. 2f . To evaluate fine-tuning in real timePlex applications, we compared the iRTs generated from DIA-NN in-silico library prediction to the empirical RTs in a LF 3-timePlex experiment. Despite there being an encoded retention time offset, the 3-timePlex data display overlap between timePlex channels. This ambiguity is due to inaccuracies in the library RTs, which we resolve through updating library RTs based on initial search results using transfer learning. Using the distribution of residuals between predicted and empirical RTs of this initial search, thresholds are dynamically set to confine the RT-space where a precursor will be considered, shown in Extended Data Fig. 4 . This fine-tuning step enhances the ability of properly assigning sample-origins to each precursor. The resulting fine-tuned and searched data produce clear temporal separation between the timePlex channels, as shown in the right panel of Fig. 2f . Benchmarking timePlex proteomic coverage through mixed-species standards To investigate the potential for false identifications in 3-timePlex, we acquired data containing H. sapiens (human) protein digests present at 10 ng in all three time channels and S. cerevisiae (yeast) digests only present at 5 ng in the last two. These data were searched with a H. sapiens , S. cerevisiae , and A. thaliana spectral library, searching for peptides from all species in each sample. At 1% FDR, we identify few yeast precursors in the time channel with missing yeast digest, and < 1% precursors correspond to A. thaliana across all three time channels, as shown in Fig. 3a . Extracted ion chromatograms were plotted for a charged peptide sequence corresponding to human GAPDH and its yeast homolog TDH1. As expected, signal was present for each human precursor across time channels, while yeast precursors had a relative absence in the Δ0 min time channel, Fig. 3b . Download figure Open in new tab Figure 3 Benchmarking timePlex protein coverage and throughput scalability a LF 3-timePlex data was acquired for human only, human & yeast, and human & yeast samples present in Δ0 min, Δ4 min, and Δ8 min timePlex channels, respectively. The data was searched with a human, yeast, and arabidopsis spectral library. The number of precursor identifications is shown in each sub-panel for each species in addition to overlapping histograms of MS1 intensities for all precursors identified. At 1% FDR 21,478, 28,030, and 16,942 human precursors, 205, 7,661, and 4,261 yeast precursors, and 112, 127, and 97 arabidopsis precursors were identified in the three timePlexes. b The total ion chromatogram (TIC) of the missing-species run is displayed in addition to extracted ion chromatograms for a human GAPDH precursor and a yeast homolog of GAPDH (TDH1). The data show signal for human precursors across all three time-channels and a relative absence of yeast signal in the Δ0 min time-channel. c Sample composition for benchmarking protein coverage and quantitative accuracy. d Sample-specific proteins identified per run at 1% FDR for LF no-plex, 3-plexDIA, LF 3-timePlex, and 3-timePlex & 3-plexDIA with the ‘separate emitters’ approach. e Protein data points identified per run at 1% FDR for LF no-plex, LF 3-timePlex, and 3-timePlex & 3-plexDIA with the ‘single emitter & joined flow’ approach. Next, we sought to benchmark the depth of coverage achieved with 3-timePlex and combinatorial 3x3 plexDIA & timePlex, relative to LF no-plex. To accomplish this, we created mixed-species standards of human and yeast proteomic digests, where human composition was held constant, and yeast was variably spiked at known differential abundances as standard practice to benchmark quantitative accuracy 50 . The sample compositions are shown in Fig. 3c . Before benchmarking timePlex to established methods, we acquired non-timePlex data with and without the timePlex setup to directly evaluate the impact of the ‘separate emitters’ timePlex implementation on ion intensities and proteome coverage. Results in Extended Data Fig. 5a-c show that the 3-column timePlex ‘separate emitters’ approach results in an ion intensity loss of approximately 2-fold and the identification of approximately 20% fewer proteins. To control for this implementation and evaluate the timePlex data-type alone, the benchmarking experiments were performed with identical LC-MS setups regardless of the methodology used for acquisition. For the implementation with separate emitters, LF no-plex quantified an average of 4,523 proteins per run between samples A, B, and C, Fig. 3d . 3-plexDIA quantified 11,224 protein data points/run and 3-timePlex quantified 11,624 protein data-points per run, corresponding to an average of 3,741 and 3,875 proteins/sample respectively. Therefore, both multiplexing approaches quantified 2.6-fold more sample-specific protein data-points/run than LF no-plex. The combination of 3x3 plexDIA & timePlex identified 25,253 sample-specific proteins per run for an average of 2,806 proteins/sample. Thus, 3x3 plexDIA & timePlex with the current implementation of multi-emitters results in 5.6-fold more sample-specific protein data-points per run, compared to LF no-plex. We observed similar proportional scaling of coverage for the single emitter (merged flow) approach with uPAC columns, Fig. 3e . With this setup, LF no-plex identified an average of 3,270 proteins between samples A, B, and C per run. 3-timePlex identified 7,919 protein data points per run, for an average of 2,640 proteins/sample, producing approximately 2.4-fold as many protein data points per run compared to LF no-plex. The combination of 3-timePlex & 3-plexDIA identified a total of 17,550 sample-specific proteins per run on average, thus approximately 1,950 sample-specific proteins per run for a 5.4-fold increase in the number of sample-specific proteins identified per run compared to LF no-plex. Search approaches that jointly model fragments, as JMod and other tools do 40 , 51 , rely on accurate spectral libraries. As the spectral complexity of DIA data increases, there may be increased sensitivity to spectral library quality. To investigate this potential, we searched 3-plexDIA and 3x3 plexDIA & timePlex data with two libraries, one optimized for mTRAQ-labeled peptides and another optimized for label-free peptides. Because the label-free fragment intensities do not accurately reflect mTRAQ-labeled peptide fragmentation, we identified 5% more precursors with the mTRAQ-optimized library than when using the label-free library, Extended Data Fig. 6a-b . Similar search of the higher plex 3x3 plexDIA & timePlex data with a mTRAQ-optimized library identified even more precursors on average, with a 14% increase. The larger increase for the higher plex data suggests that spectral library quality becomes increasingly important at increasingly high plex; this partially explains the proportionally decreased coverage per sample in 3x3 plexDIA & timePlex, which may be mitigated through improved spectral libraries. Investigating and correcting biases introduced by timePlex implementation We evaluated column-specific differences in elution profiles, shown in Extended Data Fig. 7a , as differences have the potential to impact timePlex coverage per time-channel. The columns within each implementation performed comparably, aside from one column from the ‘single emitter’ implementation which produced 2-fold broader elution peaks. As a result, the number of precursor identifications were similar within each implementation, except for the third column with the single-emitter approach, Extended Data Fig. 7c . Additionally, we observed RT-dependent differences in relative precursor ion intensities for timePlex data. Fortunately, these differences are systematic and thus correctable. This correction is conceptually similar to normalizations performed in other search software tools 15 . To stabilize relative precursor intensities across the gradient for each time channel, we applied a LOESS fit to correct relative precursor intensities with respect to retention time. Results before and after correction are shown for a single 3-timePlex run in Fig. 4a using the mixed species standards to benchmark the change in quantitative accuracy. Next, we sought to evaluate the quantitative performance of both timePlex implementations (‘separate emitters’ and ‘single emitter, merged flow’). The data indicate accurate precursor-level quantitation for both approaches, as shown in Fig. 4b . Download figure Open in new tab Figure 4 Benchmarking timePlex quantitative accuracy a Correcting changes in the relative abundances of precursor intensities across the gradient. This change is systematic and correctable as shown in quantitative assessments before and after the correction was applied. The relative intensity panels correspond to a single timePlex channel in a single run, and the quantitative ratio panels compare that timePlex channel to another channel in that same run, in this case Sample B and C are shown for yeast (orange) and human (blue) precursors. All quantitation is from MS1 precursor intensities. b Distribution density plots for MS1 ratios of human and yeast precursors in common between samples B and C ( n =9,884) are shown for both timePlex implementations to assess quantitative accuracy. Black horizontal lines mark the median of each respective distribution. c-e MS1 quantitation is shown for precursor ratios in common across all four methods and shown as scatter and box plots. The following abbreviations are used: LF-np: LF no-plex, 3-pD: 3-plexDIA, 3-tP: 3-timePlex, and 3x3: 3x3 plexDIA & timePlex. f Comparisons of MS1 and MS2-level quantitation are shown for precursor ratios in common across methods. The absolute-values of the errors between expected and observed fold-changes (FC) for each precursor ratio are shown as distributions with black vertical lines indicating the median. Benchmarking timePlex quantitative accuracy We continued to use these mixed species standards to benchmark the quantitative performance of 3-plexDIA, 3-timePlex, and combinatorial 3x3 timePlex & plexDIA to LF no-plex data using the ‘separate emitters’ approach. The number of quantified precursor ratios between sample pairs for the four methods is shown in Extended Data Fig. 8 . To fairly compare quantitative accuracy between methods, only the subset of precursor ratios quantified across all methods was used for accuracy benchmarking. All multiplexed approaches produced comparable median quantitative accuracy to LF no-plex data, but reduced precision for the 9-plex combination of 3x3 timePlex & plexDIA, as shown in Fig. 4c-e . We summarize these quantitative accuracy investigations at MS1 and MS2-levels by computing absolute errors between observed and expected fold-changes for intersected precursor ratios ( n =16,689) quantified across all methods. Quantitative accuracy remained similar across all methods at MS1 and MS2-levels, but MS1 quantitation was the most accurate in each case, Fig. 4f . Interestingly, we observed MS2 quantitation was less affected by high-plex than MS1-level quantitation. Combinatorial multiplexing of single-cell proteomes The number of samples acquired per run scales quadratically with plex when multiplexing in both mass and time domains, but linearly when implemented individually as shown in Fig. 5a . Given that the combination of 3x3 timePlex & plexDIA enables comparable quantitative accuracy at combinatorially higher throughput, we sought to evaluate its potential in maintaining high quality quantitation while scaling up the throughput of single-cell proteomics. Download figure Open in new tab Figure 5 Increasing the throughout of single-cell proteomics using combinatorial multiplexing with timePlex and plexDIA a The throughput of combinatorial plexDIA and timePlex scales quadratically, while each individually scales linearly. The dashed red line marks the throughput achieved in this implementation of 3x3 plexDIA & timePlex. b All data was acquired on an Orbitrap Exploris 240. Here, the duty cycle is illustrated, which consisted of 12 MS2 scans of variable width and an MS1 scan scheduled every 1.3 seconds. c Single-cell data from Exploris 240 acquired with 3x3 plexDIA & timePlex identified a median of 631 and 486 precursors and 295 and 242 proteins per K562 and U937 cell, respectively. d Averaged fold-changes from single-cells of K562 and U937 cell lines agrees with bulk quantitation acquired by LF 3-timePlex ( ρ =0.85). Highlighted is a protein, FABP5, found to be differentially abundant between the cell lines, and a highly abundant protein, HMGA1. e The raw MS1 extracted intensities from single-cells are plotted for a single 3x3 plexDIA & timePlex run, for precursors corresponding to HMGA1 and FABP5. The differential abundance of FABP5 is observed in the raw data of MS1 XICs, where it is higher on average in U937 cells. f PCA was performed on the subset of proteins quantified both in the bulk and single-cell samples. The bulk (diamond) and single-cell (circle) data points show agreement in their cell-type specific separation along PC1. Single cells from K562 and U937 human cell-lines were prepared by nPOP 52 , 53 using mTRAQ to isotopically label single cells to be pooled into single-cell plexDIA sets. The 3-plexDIA sets were acquired with 3-timePlex using 4 minute offsets on an Exploris 240 using a duty-cycle of 12 MS2 scans of variable length and MS1 scans interspaced every 1.3 seconds as shown in Fig. 5b . These data quantified a median of 274 proteins per cell at 1% FDR without using match between runs, Fig. 5c . To benchmark the reliability of our single-cell measurements, bulk lysate was diluted to 100-cell amounts and acquired as LF 3-timePlex. Fold-changes between cell-types indicate strong quantitative agreement between the relative protein abundances averaged across single cells and the diluted bulk lysate, with a Pearson correlation of 0.85, Fig. 5d . Among the most differentially abundant proteins which intersect with our single-cell data was FABP5, highlighted in Fig. 5d . To investigate the differential abundance of FABP5 in raw single-cell data, MS1 ion chromatograms were extracted for each single cell from a single 3x3 timePlex & plexDIA run. The total ion chromatogram (TIC) and MS1 traces are shown for each plexDIA set from each of the 4-minute separated timePlex channels. Indeed, the differential abundance of FABP5 estimated from the average across all single cells ( Fig. 5c ) is likewise reflected in raw single-cell data shown in Fig. 5d ; all estimates consistently indicate that K562 cells exhibit lower abundance of a FABP5 precursor compared to U937 cells. To verify this was not an artifact of sample preparation or acquisition, we plotted XICs for a highly abundant precursor in HMGA1 for each of the 9 single-cells, which show strong signal across all cells. Dimensionality reduction by PCA of 100x diluted bulk and single-cell data from LF 3-timePlex and 3x3 timePlex & plexDIA is shown in Fig. 5e . Single-cell 3x3 plexDIA & timePlex data clustered with 100x bulk cell lysates acquired with LF 3-timePlex. The two cell types separated along PC1, which explains 57% of the variance. To evaluate the impact of unresolved batch effects on dimensionality reduction, each sample was colored by its mass tag and the timePlex index, Extended Data Fig. 9 ; the uniform spread of tags and indexes indicated appropriate correction for any batch effects related to plexDIA and timePlex. These results highlight the potential for combinatorial multiplexing of timePlex and plexDIA to increase throughput of single-cell proteomics. 27-fold increase in proteomics throughput with 9x3 plexDIA & timePlex Recent developments in mass-tag engineering have produced tags that enhance spectral quality and enable 9-plexDIA 30 . We paired these novel mass tags (‘PSMtag’) with 3-timePlex to enable a combinatorial 27-plex. Using 25 minutes of active chromotography per time channel, we exceeded 500 samples/day. The total duration per run was 76 minutes, approximately 23.5 minutes of which was sample-loading and 52.5 minutes was data acquisition, column washing, and equilibration. We benchmarked this 27-plex with 1 ng mixed-species proteomic digests on an Orbitrap Exploris 480 with FAIMS, the compositions for which are shown in Fig. 6a . As a first evaluation of the raw data, we display the extracted ion chromatograms for select human and yeast precursors in Fig. 6b . We observed differential abundance of yeast precursors at their expected spike-in amounts, while human precursors remained similarly abundant across all mass and time channels. Download figure Open in new tab Figure 6 Combining 9-plexDIA with 3-timePlex increases proteomics throughput 27-fold a Proteomic compositions of samples A-D consisted of 1 ng human digest and variable amounts of mixed yeast digest (0.1-0.4 ng). These samples were labeled with PSMtags and pooled to enable 9-plexDIA. Samples were acquired at a rate exceeding 500 samples/day with combinatorial 9x3 plexDIA & timePlex. b Extracted ion chromatograms for yeast and human precursors from the 9x3 plexDIA & timePlex data. c Number of sample-specific proteins identified per run at 1% FDR for label-free no-plex, 9-plexDIA, and 9-plexDIA and 3-timePlex. d Bar plots show the number of protein ratios quantified across triplicates for LF no plex, 9-plexDIA, and 9x3 plexDIA & timePlex. The intersecting proteins across methods was used to benchmark quantitative accuracy at MS2-level, shown as density plots of the measured protein ratios. In this data acquisition, we again controlled for the timePlex setup and acquired triplicates of all four samples with LF no-plex, 9-plexDIA, and combinatorial 9x3 plexDIA & timePlex. The number of proteins identified at 1% FDR is shown in Fig. 6c , averaging approximately 1,200 proteins/sample for LF no-plex, 2,400 proteins/sample for 9-plexDIA, and 2,140 proteins/sample for combinatorial 9x3 plexDIA & timePlex. The lower than expected coverage for these samples likely reflects a combination of suboptimal instrument performance and the expected reduction in sensitivity from the increased total spray flow rates for the timePlex setup in its current implementation. The coverage of plexDIA samples is enhanced by the superpositions of y-ions of non-lysine terminating peptides across channels, the effect of which is especially enhanced at low input amounts. The mixed species spike-ins were used to benchmark protein-level quantitative accuracy for all methods. We observed similar, but slightly decreasing accuracy of quantitation at increasing plex Fig. 6d . These results demonstrate the ability to generate quantitative data with 27-plex DIA using combinatorial plexDIA and timePlex. Discussion Multiplexing in more than one domain (e.g. mass and time) enables combinatorially increased sample-throughput. Time-based multiplexing (‘timePlex’) was developed in this work to enable this accelerated throughput when paired with plexDIA, and was specifically demonstrated with 3-timePlex and 3-plexDIA to enable a 9-plex and 3-timePlex and 9-plexDIA to enable a 27-plex. Because multiplexing in the time-domain is label-free, the technique applies not only to bottom-up proteomics but also to other methods, such as metabolomics, where chemical labeling remains impractical or infeasible. In this work, timePlex was demonstrated with two accessible implementations which utilized separate RP-HPLC columns. However, timePlex can likely be achieved through diverse and improved means. Future implementations which utilize isocratic flow and continuous data acquisition to limit LC-associated overheads are especially promising. Capillary electrophoresis (CE) for metabolomics using multi-segment injections has achieved simultaneous acquisition of multiple samples per run through staggered and overlapped separations 37 , 54 ; as such, high-timePlex proteomics may be readily implemented through multi-segment CE approaches. Another approach could be size-exclusion chromatography (SEC) with isocratic flow as this could potentially facilitate high-timePlex from a single column. The implementation of timePlex in this work involved some drawbacks that future work can mitigate. In particular, 1) intensity variation over the course of the gradient needed correction to diminish adverse affects on quantitation ( Fig. 4a ), and 2) current implementations resulted in higher overall flow rates, which reduced sensitivity and depth of coverage. Regarding the former, differences in desolvation over the course of the gradient may result in dynamically changing ion intensities over time, among other factors which we have not fully characterized. Correcting these rapid changes becomes more difficult for time regions with sparse data. However, we found this effect remained largely correctable as observed in our benchmarking results. Regarding the latter, timePlex coverage will be improved through the use of lower flow-rates and narrower inner diameter columns to offset the time-overheads of lower flow-rates. While the implementations in their current form reduce ion intensities and thus proteomic coverage, the timePlex data-type itself, as benchmarked in this work by controlling for the implementation indicate comparable quality to non-multiplexed DIA. We expect future implementations to enable higher multiplexing both by timePlex and plexDIA that will synergistically increase throughput. For example, combining 9-plex PSMtags 30 with 9-timePlex will quadratically scale to an 81-plex, enabling the analysis of > 1,000 samples per day using DIA. Such sample throughputs have so far been attained only by using isobaric mass tags 53 , 55 , which have well characterized limitations 56 , 57 . Realizing this promise of higher-plex DIA necessitates the coordinated development of data interpretation approaches for highly parallelized mass spectra, such as JMod 40 , and continued advances in mass spectrometry instrumentation. The convergence of these advances will facilitate broader adoption of LC-MS proteomics and the generation of larger-scale accurate proteomic datasets. Methods timePlex implementation timePlex implemented in this work utilized a Dionex UltiMate 3000 HPLC. Schematics and details for sample loading are shown in Extended Data Fig. 1 . Flow from the NC pump is split using a ZenFit tee (700011518, Waters), and nanoViper™ transfer lines ranging from 20-50 µ m ID and 150 mm - 950 mm were used in combination to encode time offsets between samples. Sample-loading occured in the following way: The first sample was injected with valves in the 1 2 and 1 2 position for the first and second valves. At this stage, flow is pushed through the sample-loop into port 1 of Valve 2, which flows out of Port 2 and into Port 8 of Valve 1, and out of Port 7 to ‘Column 1’. The flows which were split from the NC pump push against plugs; as a result, all net flow is directed through the sample-loop. At this step, only ‘Column 1’ is actively receiving flow. To prevent over-pressurization, the flow rate at the step is set to 300 nL/min. Immediately after injection, the sample is run isocratically for 7 minutes with 2% ACN and 0.1% formic acid, for the sole-purpose of pushing the sample out of the loop and onto the first column. This process takes approximately 10 minutes total. The second column is loaded in a similar way, except Valve 2 is switched to the 1 10 position, directing the second sample onto ‘Column 2’. This process also takes approximately 10 minutes total. In the last 10 seconds, the flow-rate is increased to 600 nL/min in anticipation of switching the valves which occurs during the loading of the third column. The third column is loaded by switching the positions of both valves to 1 10 and 1 2 positions, respectively for the first and second valves. At this point, all three columns receive flow at a total rate from the pump of 600 nL/min. Therefore, the flow-rate across each column is approximately 200 nL/min although differences in back pressure of the columns will change the flow-rate across each column, and thus the encoded time-differential. As a result, the volume differences which encode the time-offsets may need to be adjusted to achieve the desired time-differential between timePlexes. The gradient can start immediately when loading the third sample. Additionally, because the third sample will receive its gradient through the sample-loop, which in this case holds a 1 µ L volume, it was most-efficient to have this column be the latest eluting one. All samples, including non-timePlex data acquisitions (such as LF no plex and plexDIA), were acquired with the timePlex setup so we could directly evaluate the impact on timePlex data on downstream interpretation rather than the combination of interpretation and implementation. This enabled direct evaluation of timePlex data quality. Differences between single and multi-emitter approaches Both timePlex implementations demonstrated in this work utilize the same sample-loading scheme, and a Nanospray Flex™ Ion Source on an Exploris 240. The difference between the multi-emitter (separate flow) approach and the single-emitter (joined flow) approach only pertains to how the flow is treated post-column separation, and how the ESI occurs. In the case of the multiple emitters approach, voltage (+2.2kV) was applied from the MS source to two columns via an alligator clip, and an auxiliary power supply (PS350, Stanford Research Systems) applied voltage (+2.2kV) to the third column. A PEEK cone, kindly provided by IonOpticks, held the columns in place, pointing spray at the mass-spec inlet of an Exploris 240. For the merged flow approach, flow out of each µ PAC column was merged through a 50 µ m bore cross (C360TS62, VICI Valco) to an emitter tip (TIP360002010-10-5, MS WIL). Utilizing narrow-bore crosses for merging flow post-column mitigated peak broadening. Voltage (+2.2kV) was applied to the 50 µ m bore cross via alligator clip. MS data acquisition All samples were resuspended in 0.015% n-Dodecyl β -d-maltoside (DDM) 58 with 0.1% formic acid. Each benchmarking sample, which consisted of 8 ng human digest and variable yeast digest, was acquired with the timePlex multiplexing setups shown in Fig. 2a , whether it was timePlexed or non-timePlexed. This means LF no-plex and 3-plexDIA benchmarks were acquired with the same setup as the timePlex experiments (3 columns active at a time). Single-cell data and benchmarking data for 3x3 plexDIA and timePlex were acquired on an Exploris 240 orbitrap mass spectrometer using either 25cm x 75 µ m IonOpticks columns (AUR3-25075C18) or 50 cm µ PAC™ Neo columns (COL-NANO050NEOB) with a Dionex Ultimate 3000. 8 ng LF samples were acquired in DIA mode with an MS1 scan at 120k resolving power and 300% AGC conducted every 1.3 seconds. The MS2 duty cycle consisted of 24 variable length windows coveraging 378 - 1356 m/z for LF and 478-1356 m/z for mTRAQ-labeled peptides. The MS2 windows were designed to equally distribute the ion intensities over the MS1 m/z range. Benchmarking data for 9x3 plexDIA and timePlex were acquired on an Exploris 480 orbitrap mass spectrometer with FAIMS Pro (CV=-50 and 5.2 L/min carrier gas) with 25cm x 75 µ m IonOpticks columns (AUR3-25075C18) and Dionex Ultimate 3000 HPLC. LF benchmarking For LF samples used in the 3x3 plexDIA and timePlex benchmarks, the MS2 windows were as follows: 378-387, 386.5-393, 392.5-401.5, 401-409.5, 409-418, 417.5-427.5, 427-438.5, 438-448.5, 448-460.5, 460-473.5, 473-487, 486.5-501.5, 501-517, 516.5-536.5, 536-558.5, 558-580.5, 580-607, 606.5-639, 638.5-675.5, 675-723.5, 723-784.5, 784-867, 866.5-986, and 985.5-1356 m/z . Gradient for IonOpticks columns ‘separate emitters’ approach: 600 nL/min ramping from 2% B buffer to 5.8% 2.25 minute into the gradient, 8.8% 6.3 minutes into gradient, 11.5% (11.25 min), 18.5% (31 min), 25.5% (45 min), 34.2% (54.9 min), 37.7% (56.25 min), 45% (60 min), 48% (68 min), 95% (68.2 min), 95% (72 min), 2% (72.1 min), maintaining 2% B until end (90 min). The full LC-MS method was 90 minutes. Gradient for uPAC columns ‘single emitter’ approach: 600 nL/min ramping from 2.5% B buffer to 4% 1 minute into the gradient, 8% 5.5 minutes into gradient, 15% (26 min), 18.5% (31.2 min), 25% (37.4 min), 32% (41 min), 34% (51.5 min), 95% (52 min), 95% (58 min), 2% (64 min), maintaining 2% B until end (75 min), with the last 10 minutes run at 1,200 nL/min. The full LC-MS method was 75 minutes. For LF samples used in the 9x3 plexDIA and timePlex benchmarks, MS1 scan was performed at 180k resolving power every 1.3 seconds (500% AGC), and MS2 scans were performed at 45k resolving power (1000% AGC) with the following windows: 380-426.8, 426.3-461.5, 461-491.3, 490.8-517.5, 517-544.9, 544.4-565, 564.5-582.6, 582.1-596.6, 596.1-610.6, 610.1-629.5, 629-649.5, 649-664.7, 664.2-683.6, 683.1-697.6, 697.1-718.9, 718.4-741.4, 740.9-774.9, 774.5-813.2, 812.7-869.8, 869.3-1000 m/z . The following gradient was used for acquiring the LF data shown in the 9x3 plexDIA and timePlex benchmarks: 2% B to 5% in 1 minute at 600 nl/min, ramping to 26% B by 20.4 minutes, then 35% B by 25 minutes, 37% B by 31.5 minutes, and the wash was performing by ramping to 95% B by 32 minutes holding until minute 37. The B buffer was lowered to 2% B by minute 37.1, held isocratically until minute 47.1 where the flow rate was increased to 1 µ L/min until minute 50, after which the flow rate was reduced to 300 nL/min until the end (52.5 minutes). mTRAQ-labeled benchmarks For mTRAQ-labeled benchmarks, the MS2 windows were as follows: 478-485.5, 485-493, 492.5-501.5, 501-510.5, 510-518, 517.5-526.5, 526-535.5, 535-544.5, 544-553, 552.5-563, 56 2.5-574.5, 574-586.5, 586-598.5, 598-612.5, 612-627, 626.5-643.5, 643-662, 661.5-683, 682.5-707.5, 707-739, 738.5-778, 777.5-834, 833.5-940, and 939.5-1356 m/z . Gradient for IonOpticks columns ‘separate emitters’ approach: 600 nL/min ramping from 2% B buffer to 6.5% 2.25 minute into the gradient, 9.5% 6.3 minutes into gradient, 12% (11.25 min), 19.4% (31 min), 26.4% (45 min), 35.1% (54.9 min), 46% (60 min), 49% (68 min), 95% (68.2 min), 95% (72 min), 2% (72.1 min), maintaining 2% B until end (90 min). The full LC-MS method was 90 minutes. Gradient for uPAC columns ‘single emitter’ approach: 600 nL/min ramping from 4% B buffer to 8% 5.5 minute into the gradient, 16% 26 minutes into gradient, 15% (26 min), 19.5% (31.2 min), 26% (37.4 min), 33.3% (41 min), 36.5% (51.5 min), 95% (52 min), 95% (58 min), 2% (58.1 min), maintaining 2% B until end (75 min), with the last 10 minutes run at 1,200 nL/min. The full LC-MS method was 76 minutes. PSMtag-labeled benchmarks For PSMtag-labeled benchmarks, FAIMS was utilized with CV = -50 with an MS1 scan at 180k resolving power (500% AGC) performed every 1.3 seconds. The MS2 scans were performed at 45k (1000% AGC) resolving power with the following windows: 500-538.5, 538-566.5, 566-590.5, 590-611.5, 611-633.5, 633-649.5, 649-663.5, 663-674.5, 674-685.5, 685-700.5, 700-716.5, 716-728.5, 728-743.5, 743-754.5, 754-771.5, 771-789.5, 789-816.5, 816-847.5, 847-893.5, 893-1000 m/z . The gradient was as follows: 600 nL/min ramping from 24% B buffer to 30.4% 0.15 minutes into the gradient, 49.6% 20.4 minutes into gradient, 60% (25 min), 62% (31.5 min), 95% (32 min), 95% (37 min), 24% (37.1 min), and at 47.1 min the flow rate is increased to 1 µ L/min until 50 minutes, and then lowered to 300 nL/min until the end (52.5 minutes). plexDIA and timePlex single-cells For plexDIA and timePlex single-cell acquisition, the MS2 windows were as follows: 478-493, 492.5-510.5, 510-526.5, 526-544.5, 544-563, 562.5-586.5, 586-612.5, 612-643.5, 643-683, 682.5-739, 738.5-834, and 833.5-1356 m/z . Gradient for IonOpticks columns ‘separate emitters’ approach: 600 nL/min ramping from 2% B buffer to 6.5% 1.5 minute into the gradient, 9.5% 4.16 minutes into gradient, 12% (7.425 min), 19.4% (20.46 min), 26.4% (29.7 min), 35.1% (36.6 min), 46% (40 min), 49% (48 min), 95% (48.2 min), 95% (52 min), 2% (52.1 min), maintaining 2% B until end (70 min). Spectral library generation with DIA-NN Data was generated to create empirical libraries for searching the human and yeast benchmarking data. The library-generating data consisted of 250 ng injections of label-free and mTRAQ-labeled human (K562) and yeast standards, acquired on the Exploris 240 using 25cm x 75 µ m IonOpticks columns (AUR3-25075C18) and a Dionex UltiMate 3000 HPLC. The data was searched with DIA-NN version 1.9 15 using a predicted spectral library. To generate an empirical library for searching single-cells, 8 ng injections of mTRAQ-labeled K562 and U937 protein digests were similarly searched. Retention time prediction and transfer learning Empirical libraries output from DIA-NN 15 , which were generated from label-free and mTRAQ-labeled data, were used to train a model to predict peptide iRTs. Precursors from the empirical libraries were collapsed to peptide-level, averaging the iRT between charge-states, if there were any reported differences. The peptides were one-hot encoded as a 30x20 array for each peptide, where 30 is the maximum length for a peptide and 20 is the number of possible amino acids. A hybrid convolutional and fully connected feedforward neural network was trained to predict peptide RTs, using TensorFlow. The model consisted of four 1D convolutional layers, with the number of filters doubling at each successive layer: 32, 64, 128, and 256. Each layer used a kernel size of 4. No pooling was applied after the first two convolutional layers. Max pooling with a pool size of 2 was applied after each of the final two convolutional layers. The output of the final convolutional layer was flattened and passed into a fully connected feedforward neural network with five layers consisting of 1000, 256, 128, 64, and 32 nodes, respectively. All layers used the ‘Swish’ activation function 59 , except for the final output layer, which used a linear activation to produce the final RT value. An ensemble of five models was created and used for transfer learning; predictions were the average of the five model outputs. The models were trained with early stopping (monitor=’val loss’, patience = 20, restore best weights=True), epochs = 150, batch size = 32, validation split = 0.1, with Adam 60 as the optimizer (learning rate=0.001) and mean absolute error as the loss function. Transfer learning was applied to fine-tune library RTs. This involved re-initializing the five models and training on new data stemming from the initial search which occurs in JMod for m/z and RT alignment. This initial search seeks to identify peptides based on the peptide-like features output by Biosaur2 61 . These features are only used for the initial search to update m/z and RT alignments, and refine library RTs. The transfer learning occurs under the following parameters: early stopping was enabled (monitor=’val loss’, patience = 10, restore best weights=True), epochs = 50, batch size=24, optimizer=’Adam’ (learning rate=0.0001), loss = ‘mae’, validation split = 0.1. The average RT output across the five models is used as the final RT prediction in the library. Finally, the RT predictions from transfer learning are compared to the original library RTs using the validation set of data. The strategy with the greatest RT accuracy is used for the full-search in JMod. Searching with JMod JMod is an open source proteomics search software, which leverages joint-modeling of spectra to enable interpretation of proteomics data acquired by DIA 40 . It begins by identifying peptidelike features which are output by feature detection software 61 . JMod then attempts to identify the precursor that best fits each of these features individually. This first search does not utilize any retention time information and, due to JMod’s approach, can identify multiple instances of the same precursor. For n -timePlex, precursors identified n times are extracted. These are assumed to have one identification for each column. The time differences encoded by the timePlex setup remain relatively constant across the gradient. Therefore, correctly matched precursor sets will all have systematic offsets. Importantly, this does not require the same offset to be encoded between different pairs of successive columns. However, the consistency between each pair of columns allows us to further filter our set of precursors to those that agree on the time encoding. This provides a set of high-confidence identifications across the gradient which are separated into different timePlexes. A LOWESS regression is fit to each individual timePlex channel to align the library retention times to the observed data. The first timePlex channel is also used as the reference set of identifications for retention time prediction. The peptides sequence and retention time pairs are used to finetune a retention time prediction model. LOWESS regressions are then also fit for the finetuned predictions vs the observed retention times for each timePlex. The differences between these fits and the observed retention times are used to calculate retention time errors. Cumulative distributions are calculated for both the original library retention time errors and the newly predicted retention time errors. The cumulative distribution with the largest area under the curve is selected as the best fit Extended Data Fig. 4 . This is the model that minimizes errors. Then, to calculate the optimal retention time tolerance, both exponential and Gaussian cumulative distributions are compared to the observed cumulative distribution. The distribution that matches the observed data best is used to calculate the 95th percentile which will specify the retention time tolerance. Once a best model is selected, JMod duplicates the library for n -timePlex. Each will have an updated retention time based on the retention times and respective LOWESS fit of the best approach identified previously. These tolerances are also made to be non-overlapping between timePlex channels. This ensures that the same precursor in each channel will be considered independent of the other channels, leading to column and sample-specific identifications. Once the library is created, the data are searched the same as any other experiment. Human and yeast benchmarking LF no-plex human and yeast benchmarking data was searched with JMod (v0.1.0 beta) using the following commands for the set of data used in the 3x3 plexDIA and timePlex comparisons: {– iso}, {–num iso 2}, {–lib frac 0.5}. LF 3-timePlex data was searched using the following commands: {–iso}, {–num iso 2}, {–lib frac 0.5}, {–timeplex}, {–num timeplex 3}. LF data used in the 9x3 plexDIA and timePlex benchmarks were analyzed with JMod (v0.1.0) with the following commands: {–iso}, {–num iso 3}, {–ppm 10} {–ms1 ppm 10} {–lib frac 0.2} {–score lib frac 0.2}, {–atleast m 1}. The 3-plexDIA human and yeast benchmarking data was searched with JMod (v0.1.0 beta) using the following commands: {–iso}, {–num iso 2}, {–plexDIA}, {–tag mTRAQ}, {–lib frac 0.5}. 9-plexDIA data was searched with JMod (v0.1.0) with the following commands: {–iso}, {–num iso 3}, {–plexDIA}, {–ppm 10}, {–ms1 ppm 10}, {–lib frac 0.3}, {–score lib frac 0.3}, {–tag tag6 9plex}, {–use emp rt}. Combinatorial 3-timePlex and 3-plexDIA human and yeast benchmarking data was searched with JMod (v0.1.0 beta) using the following commands: {–iso}, {–num iso 2}, {–plexDIA}, {–tag mTRAQ}, {–lib frac 0.5}, {–timeplex}, {–num timeplex 3}. Combinatorial 3-timePlex and 9-plexDIA data was searched with JMod (v0.1.0) with the following commands: {–iso}, {– num iso 3}, {–plexDIA}, {–ppm 10}, {–ms1 ppm 10}, {–lib frac 0.3}, {–score lib frac 0.3}, {–tag tag6 9plex}, {–use emp rt}, {–timeplex}, {–num timeplex 3}. Because the empirical library was generated using IonOpticks columns at the same time the samples were generated, there was no need for RT fine-tuning on the benchmarking samples which used IonOpticks columns. In these cases, we additionally used the following command {–use emp rt} to use the original library RTs for searching rather than performing RT fine-tuning. However, fine-tuning was applied for benchmarking data acquired with µ PAC columns, as the library RTs were unsuitable. All non-multiplexed data and FDR benchmarking data was filtered at 1% precursor and protein FDR (Qvalue < 0.01 & Protein Qvalue < 0.01). Less confident precursor assignments were allowed for multiplexed data if at least one channel in the run identified the same precursor at 1% FDR, which we refer to as BestChannel Qvalue. In these cases, the following filters were applied: BestChannel Qvalue < 0.01 & Qvalue < 0.2 & Protein Qvalue < 0.01. Searching the single-cell spectra Single cell combinatorial 3-timePlex and 3-plexDIA data was searched with JMod using an empirical spectral library generated from 8 ng of K562 and U937 cell digests using the following commands: {–iso}, {–num iso 2}, {–plexDIA}, {–tag mTRAQ}, {–lib frac 0.5}, {–timeplex}, {–num timeplex 3}, {–lib frac 0.20}, {–score lib frac 0.33}, {–user rt tol}, {–rt tol 0.45}, {– initial percentile 20}, {–user percentile}. Sample preparation Creating a mixed species standard for benchmarking Benchmarking samples consisted of commercial Yeast (V7461, Promega) and Human K562 (V6951, Promega) protein digests. The dried digests were resuspended in 100 mM TEAB, pooled using their appropriate ratios to create Samples A, B, and C, each sample consisted of 25 µ g of total human material for each sample. The samples were split in two, one to remain label-free and the other to be chemically labeled with mTRAQ. The label-free samples were dried, and the other set of samples were labeled with mTRAQ Δ0 (SCIEX 4440015), Δ4 (SCIEX 4427698), and Δ8 (SCIEX 4427700) at a 2:1 ratio of 500 ng/uL sample in 100 mM TEAB : stock concentration of tag. After addition of the label, the samples were left at room temperature to react for 2 hours. Hydroxylamine was added to a final concentration of 0.2% to quench the reaction at room temperature for 30 minutes. The samples were then dried with a speedvacuum, and then resuspended in 0.015% DDM and 0.1% formic acid to their appropriate concentrations. PSMtag benchmarks were prepared and reconstituted as described by Specht et al 30 . Cell culture K562 (CCL-243, ATCC) and U937 (CRL-1593.2, ATCC) cells were cultured in RPMI-1640 Medium (R8758, Sigma-Aldrich) supplemented with 10% fetal bovine serum (10439016, Gibco) and 1% penicillin-streptomycin (15140122, Gibco), grown at 37 ◦ C and 5% CO 2 , and harvested within 15 passes. 100x bulk K562 and U937 K562 and U937 human cells were harvested from culture, washed 2x in PBS, and resuspended in water to a concentration of approximately 2000 cells/ µ L. The cell suspensions were subjected to mPOP 62 . In brief, the cell suspensions were frozen at -80°C, then heated at 90°C in a thermalcycler for 10 minutes. TEAB was then added to the cell lysate to a concentration of 100 mM, Benzonase nuclease (E1014, Millipore) was added to 0.1 U/ µ L, and Trypsin Platinum (VA9000, Promega) was added to a concentration of 15 ng/ µ L. Single-cell sample preparation using nPOP Single cells were prepared for proteomic analysis by nPOP as previously described by Leduc et al. 52 , 53 . In short, single-cells (in a volume of 0.3 nL) were sorted by CellenOne into 9 nL droplets of 100% DMSO on a fluorocarbon-coated glass slide to lyse. The lysed cells were digested with a mixture containing Trypsin Platinum (VA9000, Promega) at a concentration of 120 ng/ µ L and 0.015% DDM (w/v) and 5 mM HEPES (pH 8.5), 13.5 nL of digestion mixture was dispensed onto each droplet. Peptides were then chemically labeled with mTRAQ Δ0, Δ4, or Δ8 (20 nL/droplet) and subsequent addition of 100 mM TEAB buffer (13.5 nL/ droplets, pH 8.5) was added for a 1 hr incubation period. Single cells were automatically pooled by the CellenOne into 3-plex sets, and dispensed into wells of a 384 well-plate. The pooled samples in the 384 well-plate were dried in a speed-vacuum then stored at -80°C until use. Computational data analysis Quantification For 3x3 plexDIA and timePlex benchmarking experiments, precursors were quantified by MS1 Area for LF no-plex and LF 3-timePlex data, or by a custom algorithm for 3-plexDIA and combinatorial 3-plexDIA & 3-timePlex acquisitions. The algorithm identifies the best fitting MS1 scan across the channels in a plexDIA set by computing a correlation of the theoretical isotopic envelopes of the precursor group to the empirically observed envelope. The MS1 scan with the highest correlation was considered the MS1 apex. Quantities at this scan were extracted for single-cell data; for bulk data, this scan and adjacent scans (radius = 1) were summed for MS1 quantitation. For 9x3 plexDIA and timePlex benchmarking experiments precursors were quantified at MS2-level using ‘coeff’ for all data. plexDIA data and combinatorial 9x3 plexDIA and timeplex were filtered such that non-lysine C-terminating peptides quantified at least 2 b-ions and were required to have channel specific Q-value < 0.05. Protein-level quantities were computed with MaxLFQ 63 . All quantitative benchmarking was performed using the median value across triplicate samples. For example, in the case of LF no-plex 9 LC-MS runs were required to analyze triplicates of Samples A, B, and C, respectively; 3 runs are required to analyze triplicates of the same samples for 3-plexDIA or 3-timePlex; and finally, a single run was used to analyze triplicates of the same samples for 3x3 plexDIA & timePlex. This way, medians were computed from the same number of samples regardless of the the method. This was performed for all quantitative comparisons unless otherwise specified. Correcting RT biases in relative quantitation Precursor quantities were normalized to account for sample-loading differences by dividing each sample by its respective median, then relative quantities for each precursor were computed by dividing each normalized value by the mean across samples. In the case of mixed-species spikeins, the resulting data were filtered for human precursors only. These relative values were fit with a LOESS regression as a function of RT, and the fit was subtracted from the relative values to correct them. These corrected values were used for all downstream analyses. Computing full width half max (FWHM) of precursor elution profiles Outputs from JMod list the sequence of MS1 intensities for each precursor. Using this list, the FWHM for each precursor was computed by fitting a gaussian of their MS1 elution profiles. The interpolated apex and surrounding interpolated profile was used to find the range (in number of scans) where the profile was at half the maximum. Given that the duty cycle was scheduled to perform an MS1-scan every 1.3 seconds, the FWHM in scans was multiplied by this scan rate to compute the final FWHM in seconds. Entrapment to assess false discoveries of timePlex An empirical library generated by DIA-NN 15 for human and yeast LF peptides was created, as a well as a predicted library for A. thaliana . Precursors from the A. thaliana predicted library were randomly sampled to have the same total number as in the empirical human and yeast library. The two equivalently sized libraries were merged, and this merged library was used to search LF 3-timePlex data which had missing yeast in one time channel, and A. thaliana missing from all time channels. Data were analyzed at 1% FDR. Single-cell protein quantification, imputation, batch correction, and PCA To remove cells which failed during preparation or acquisition, single-cells were excluded from downstream analyses if they quantified fewer than 250 precursors. Precursors were quantified at MS1-level at the elution apex only. Precursor abundances were corrected for RT-dependent changes in relative abundances as described above. These relative values were then scaled by the mean of the original values across cells to scale them back into their original intensity ranges. These values were then used for MaxLFQ quantitation using the DIANN R package 15 , 63 . Protein quantities were normalized by their median abundances for each cell to account for absolute differences, then normalized across cells by the mean. The log 2 -transformed data was then imputed using a weighted kNN approach (k=5), where the weights were scaled based on euclidean distances to its nearest neighbors. And finally, the data was batch-corrected using ComBat 64 , 65 to account for mTRAQ label and column biases. Single cell and 100x bulk data were filtered to have intersected proteins, and principal components were computed from imputed data where the variance of each protein was scaled by the sum of squared protein-protein correlations. Evaluating agreement between single-cell and bulk timePlex quantitation Precursor quantities were filtered for 1% FDR for bulk while allowing additional precursor identifications for single-cells if at least one channel in the run passed 1% FDR (BestChannel Qvalue < 0.01). The minimum protein q-value across the single-cell data set was used to filter proteins at 1% FDR. All relative precursor abundances were normalized to resolve retention-time associated biases in precursor intensities. The median relative values computed across single-cells was used for each protein, requiring at least 20% data-completeness for a protein to be included in the comparison relative to the 100x bulk. A Pearson correlation was then computed for the averaged single-cell fold-changes to the averaged 100x bulk fold-changes. Differential protein abundance analysis Precursor quantities were filtered for Qvalue < 0.01 and Protein QValue < 0.01 for all data. The dynamic changes in relative precursor intensities which change as a function of retention time were normalized as previously described. These corrected precursors intensities were used for differential protein abundance analysis calculated through the MS-Empire workflow 66 . In total, there were 6 replicates for each cell-type, two from each column. Precursors used for analysis were required to be quantified in at least two replicates per condition. Data normalization was performed using default MS-Empire workflow operations, as described by Ammar et al. 66 . Data and code availability Raw data, spectral libraries, search outputs, and meta data are available at MassIVE: MSV000097736. Code for repeating downstream analyses is available at github.com/ParallelSquared/timePlex. Competing interests J.D., H.S., K.M.D., and N.S. are listed as inventors on a provisional patent application for timePlex. N.S. is a founding director and CEO of Parallel Squared Technology Institute, which is a nonprofit research institute. Author contributions Experimental design : J.D., H.S., and N.S. Data analysis : J.D., K.M.D., and N.W. LC-MS/MS : J.D. Sample preparation : J.D., P.S., and M.Y. Raising funding : N.S. and H.S. Supervision : N.S. Initial draft : J.D., and N.S. Writing : All authors approved the final manuscript. Extended Data Figures Download figure Open in new tab Extended Data Fig. 1 Implementing 3-timePlex on Dionex UltiMate 3000 HPLC. The diagram illustrates how samples are loaded sequentially for each column and then run together. Red indicates where net flow occurs. Flow is bifurcated from the NC pump, one line leads to the sample loop to push sample onto a given column, and the other line is bifurcated again to ports 1 and 6 of Valve 1. Plugs are utilized to maintain pressure to push sample through the loop during column loading. The valve-positions required to achieve this are noted on each panel of the diagram. Flow-rate of the methods are adjusted to account for the number of columns actively receiving flow. In this implementation, columns 1 and 2 are loaded at 300 nL/min, where only a single column is active a a time; this produces approximately 1.5x the backpressure as three columns run at 600 nL/min, which occurs during the ‘Load column 3 and run all columns’ phase. Download figure Open in new tab Extended Data Fig. 2 Assessing carryover between timePlex channels Sample loading occurs through a shared sample-loop and transfer lines. In some cases, this may result in unspecific binding of peptides onto these surfaces which will flow (unintended) into the third column during the active gradient. To mitigate this, samples were resuspended in 0.015% DDM for sample-loading. To assess the carryover, LF non-multiplexed data were searched as a 2-timePlex, attempting to quantify carryover (if any exists) in the unintended time channel. If signal was only detected in the intended time channel, that ratio was given a separable value of 1e-6 (green), if it was only detected in the unintended channel the ratio was given a value of 1e6 (blue), otherwise the observed ratio was plotted (red). Log10-transformed ratios reflect mostly missing data in the unintended time channel, and low signal for non-missing values. The median intensity of the non-missing data was 5.7% of the intensity of the intended time-channel. Download figure Open in new tab Extended Data Fig. 3 Model architecture for predicting RTs The model is a hybrid CNN and fully connected feedforward DNN which takes inputs of shape (30, 20), 30 being the peptide length and 20 being placeholders for each amino acid. There are 4 1D convolutional layers with kernel size = 4. The number of filters applied at each convolutional layer was set to double, starting at 32 and ending with 256. Max pooling = 2 was used for the final two layers. The final convolutional layer was flattened and fed into a fully connected feed-forward deep neural network with 1000, 256, 128, 64, and 32 nodes at each layer. Swish activation functions were used at all layers, with the exception of the final layer which was linear for computing the final RT output. Download figure Open in new tab Extended Data Fig. 4 Setting RT tolerances JMod searches benefit from thresholding on a tightly set RT range, so a peptide is only considered within a specific RT window and not throughout the run. This is especially important in timePlex, where peptides must be correctly associated to their sample origin. To set appropriate RT windows, we compute a cumulative distribution on the validation dataset, comparing the original library RTs with the fine-tuned RTs. We fit the cumulative distributions as Gaussian or exponential, and choose the fit that best matches the actual distribution, setting the RT threshold at the 95th percentile. a Fine-tuning is not necessary for data where the library has accurate RTs, as is the case here where the libraries were acquired shortly after the benchmarking runs and originate from the same columns used in the benchmarking data. b However, in cases where the library RTs do not properly model the empirical RTs, fine-tuning (orange) is highly beneficial, allowing for a smaller range to consider a peptide eluting. Here the library originates from RTs from IonOpticks columns, leading to a large discrepancy between what is empirically observed on µ PAC columns (blue). Download figure Open in new tab Extended Data Fig. 5 Investigating the effect of the 3 column separate emitter approach (background spray) Here we investigate how the 3-column setup alone (not necessarily timePlex) impacts ion intensities and coverage relative to standard proteomics acquisition from a single emitter. The flow rate for the 3 column approach is 600 nL/min (approximately 200 nL/min/column) or 200 nL/min for a 1 column setup. a Ratio of precursor MS1 intenisites for the same samples acquired when 3 columns are spraying a total of 600 nL/min (however, only 1 column is actively separating a sample), or 1 column is spraying 200 nL/min. The 3-column setup results in approximately 2-fold lower MS1 intensities. The red dashed line indicates the median of the distribution. b We investigate how the 3-column setup impacts precursors identified for non-timePlex applications. The data reflect a 25% reduction in precursors identified with the 3-column setup, and c an approximately 20% reduction in protein coverage. Download figure Open in new tab Extended Data Fig. 6 Spectral library quality becomes increasingly important at high-plex a Triplicates of each plexDIA or combinatorial plexDIA & timePlex set were acquired, resulting in 9 runs for 3-plexDIA and 3 runs for 3x3 plexDIA & timePlex. Shown is the number of precursors identified for each run, searched with an unoptimized library (blue) or mTRAQ optimized library (red). b The same data, but shown as the change in the number of precursors identified as a result of searching data with an unoptimized library or mTRAQ optimized library. Searching data with an optimized spectral library results in a greater percentage increase of identifications at higher plex. Download figure Open in new tab Extended Data Fig. 7 Investigating column-specific differences with timePlex a Full width half max (FWHM) of precursor elutions was intersected across columns and implementations for the ‘separate emitters’ approach using IonOpticks 25cm x 75 µ m columns, or the ‘single emitter’ approach using µ PAC Neo 50 cm columns, and plotted as density plots; a vertical black line marks the median of the distributions. b The ‘separate emitters’ approach using IonOpticks columns produced similar precursor coverage across columns. In the figure, each point represents a replicate and each facet corresponds to a samples (A, B, C). c Similar trends are observed for the ‘single-emitter joined flow’ approach using Thermo µ PAC Neo columns, however the the 3rd column consistently under-performed the first two. Download figure Open in new tab Extended Data Fig. 8 Precursor ratio counts between human and yeast benchmarking samples a-c Number of precursor ratios quantified for each method between samples A/B, A/C, and B/C. Download figure Open in new tab Extended Data Fig. 9 Investigating the presence of mass-tag and timePlex batch effects in PCA after batch correction a PCA of single-cells and bulk samples colored by the mass tag label with which they were acquired. b PCA of single-cells and bulk samples colored by the timePlex channel with which they were acquired. This also corresponds to the LC column which was used. Acknowledgments We thank Sarah Sipe for discussions regarding timePlex implementation and Jarrod Sandow (IonOpticks) for equipment and suggestions for implementing the ‘separateemitter’ approach in timePlex. PTI is a Convergent Research Focused Research Organization (FRO) and has received support from Eric and Wendy Schmidt as well as Griffin Catalyst. Initial conceptualization occurred when J.D. and N.S. were supported by a MIRA award from the NIGMS of the NIH (R35GM148218) to N.S. Funder Information Declared NIH , R35GM148218 Footnotes ∈ Data, code & protocols: parallelsq.org/timePlex https://www.parallelsq.org/timeplex https://github.com/ParallelSquared/timePlex References 1. ↵ Demichev , V. et al. dia-PASEF data analysis using FragPipe and DIA-NN for deep proteomics of low sample amounts . en. Nature Communications 13 , 3944 . ( 2024 ) ( July 2022). OpenUrl 2. ↵ Kreimer , S. et al. Parallelization with Dual-Trap Single-Column Configuration Maximizes Throughput of Proteomic Analysis . Analytical chemistry 94 , 12452 – 12460 ( Sept . 2022 ). OpenUrl CrossRef 3. ↵ Xie , X. et al. Multicolumn Nanoflow Liquid Chromatography with Accelerated Offline Gradient Generation for Robust and Sensitive Single-Cell Proteome Profiling . Analytical Chemistry 96 , 10534 – 10542 ( July 2024 ). OpenUrl CrossRef 4. ↵ Nagae , N. , Itoh , H. , Nimura , N. , Kinoshita , T. & Takeuchi , T . Rapid separation of proteins and peptides by reversed-phase microcolumn liquid chromatography . Journal of Microcolumn Separations 3 , 5 – 9 ( 1991 ). OpenUrl CrossRef 5. Moritz , R. L. & Simpson , R. J . Purification of proteins and peptides for sequence analysis using microcolumn liquid chromatography . Journal of Microcolumn Separations 4 , 485 – 489 ( 1992 ). OpenUrl CrossRef 6. Xu , P. , Duong , D. M. & Peng , J . Systematical Optimization of Reverse-Phase Chromatography for Shotgun Proteomics . Journal of Proteome Research 8 , 3944 – 3950 ( Aug . 2009 ). OpenUrl CrossRef PubMed Web of Science 7. ↵ Žuvela , P. , et al. Column Characterization and Selection Systems in Reversed-Phase High-Performance Liquid Chromatography . Chemical Reviews 119 , 3674 – 3729 ( Mar . 2019 ). OpenUrl CrossRef PubMed 8. ↵ Michalski , A. et al. Mass Spectrometry-based Proteomics Using Q Exactive, a High-performance Benchtop Quadrupole Orbitrap Mass Spectrometer * . Molecular & Cellular Proteomics 10 ( Sept . 2011 ). 9. Meier , F. et al. Parallel Accumulation–Serial Fragmentation (PASEF): Multiplying Sequencing Speed and Sensitivity by Synchronized Scans in a Trapped Ion Mobility Device . Journal of Proteome Research 14 , 5378 – 5387 . ( 2022 ) ( Dec . 2015). OpenUrl 10. Meier , F. et al. Online Parallel Accumulation-Serial Fragmentation (PASEF) with a Novel Trapped Ion Mobility Mass Spectrometer . Molecular & cellular proteomics: MCP 17 , 2534 – 2545 ( Dec . 2018 ). OpenUrl CrossRef PubMed 11. Stewart , H. I. et al. Parallelized Acquisition of Orbitrap and Astral Analyzers Enables High-Throughput Quantitative Analysis . Analytical Chemistry 95 , 15656 – 15664 ( Oct . 2023 ). OpenUrl CrossRef 12. Huffman , R. G. et al. Prioritized mass spectrometry increases the depth, sensitivity and data completeness of single-cell proteomics . Nat. Methods ( 2023 ). 13. ↵ Peters-Clarke , T. M. , Coon , J. J. & Riley , N. M . Instrumentation at the leading edge of proteomics . Analytical Chemistry 96 , 7976 – 8010 ( 2024 ). OpenUrl CrossRef 14. ↵ Tsou , C.-C. et al. DIA-Umpire: comprehensive computational framework for data-independent acquisition proteomics . Nature methods 12 , 258 – 264 ( 2015 ). OpenUrl CrossRef PubMed 15. ↵ Demichev , V. , Messner , C. B. , Vernardis , S. I. , Lilley , K. S. & Ralser , M . DIA-NN: Neural networks and interference correction enable deep proteome coverage in high throughput . Nature methods 17 , 41 – 44 . ( 2022 ) ( Jan . 2020). OpenUrl 16. Cox , J . Prediction of peptide mass spectral libraries with machine learning . Nature Biotechnology 41 , 33 – 43 ( 2023 ). OpenUrl CrossRef PubMed 17. ↵ Slavov , N . Driving Single Cell Proteomics Forward with Innovation . Journal of Proteome Research 20 , 4915 – 4918 ( 2021 ). OpenUrl CrossRef PubMed 18. ↵ Bache , N. et al. A Novel LC System Embeds Analytes in Pre-formed Gradients for Rapid, Ultra-robust Proteomics * . Molecular & Cellular Proteomics 17 , 2284 – 2296 ( Nov . 2018 ). OpenUrl CrossRef PubMed 19. ↵ Zahn , E. et al. Development of a High-Throughput Platform for Quantitation of Histone Modifications on a New QTOF Instrument . Molecular & Cellular Proteomics 24 ( Jan. 2025 ). 20. ↵ Derks , J. et al. Increasing the throughput of sensitive proteomics by plexDIA . en. Nat. Biotechnol. ( July 2022 ). 21. ↵ Thielert , M. et al. Robust dimethyl-based multiplex-DIA doubles single-cell proteome depth via a reference channel . Molecular Systems Biology 19 , e11503 . ( 2024 ) ( Sept . 2023). OpenUrl 22. ↵ Orsburn , B. C . Single cell proteomics by mass spectrometry reveals deep epigenetic insight into the actions of an orphan histone deacetylase inhibitor . bioRxiv: The Preprint Server for Biology , 2024.01.05.574437 ( June 2024 ). 23. Derks , J. , et al. Single-nucleus proteomics identifies regulators of protein transport . bioRxiv: The Preprint Server for Biology , 2024.06.17.599449 ( June 2024 ). 24. ↵ Slavov , N . Scaling Up Single-Cell Proteomics . Molecular & Cellular Proteomics 21 , 100179 ( 2022 ). OpenUrl CrossRef PubMed 25. ↵ Gygi , S. P. et al. Quantitative analysis of complex protein mixtures using isotope-coded affinity tags . Nature Biotechnology 17 , 994 – 999 ( Oct . 1999 ). OpenUrl CrossRef PubMed Web of Science 26. Blagoev , B. , Ong , S.-E. , Kratchmarova , I. & Mann , M . Temporal analysis of phosphotyrosine-dependent signaling networks by quantitative proteomics . Nature Biotechnology 22 , 1139 – 1145 ( Sept . 2004 ). OpenUrl CrossRef PubMed Web of Science 27. Boersema , P. J. , Raijmakers , R. , Lemeer , S. , Mohammed , S. & Heck , A. J. R . Multiplex peptide stable isotope dimethyl labeling for quantitative proteomics . Nature Protocols 4 , 484 – 494 ( Apr . 2009 ). OpenUrl CrossRef PubMed 28. DeSouza , L. V. et al. Multiple reaction monitoring of mTRAQ-labeled peptides enables absolute quantification of endogenous levels of a potential cancer marker in cancerous and normal endometrial tissues . en. J. Proteome Res . 7 , 3525 – 3534 ( Aug . 2008 ). OpenUrl CrossRef PubMed Web of Science 29. DeSouza , L. V. et al. mTRAQ-based quantification of potential endometrial carcinoma biomarkers from archived formalin-fixed paraffin-embedded tissues . PROTEOMICS 10 , 3108 – 3116 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 30. ↵ Specht , H. , et al. PSMtags improve peptide sequencing and throughput in sensitive proteomics . bioRxiv ( 2025 ). 31. ↵ Thompson , A. et al. Tandem Mass Tags: A Novel Quantification Strategy for Comparative Analysis of Complex Protein Mixtures by MS/MS . Analytical Chemistry 75 , 1895 – 1904 ( Apr . 2003 ). OpenUrl CrossRef PubMed 32. ↵ Wiese , S. , Reidegeld , K. A. , Meyer , H. E. & Warscheid , B . Protein labeling by iTRAQ: a new tool for quantitative mass spectrometry in proteome research . eng. Proteomics 7 , 340 – 350 ( Feb . 2007 ). OpenUrl CrossRef PubMed Web of Science 33. ↵ Zuniga , N. R. et al. Achieving a 35-Plex Tandem Mass Tag Reagent Set through Deuterium Incorporation . Journal of Proteome Research 23 , 5153 – 5165 ( Nov . 2024 ). OpenUrl CrossRef PubMed 34. ↵ Venable , J. D. , Dong , M.-Q. , Wohlschlegel , J. , Dillin , A. & Yates , J. R . Automated approach for quantitative analysis of complex peptide mixtures from tandem mass spectra . en. Nature Methods 1 , 39 – 45 . ( 2020 ) ( Oct . 2004). OpenUrl 35. ↵ Ludwig , C. et al. Data-independent acquisition-based SWATH-MS for quantitative proteomics: a tutorial . Molecular Systems Biology 14 , e8126 ( Aug. 2018 ). OpenUrl Abstract / FREE Full Text 36. ↵ Derks , J. & Slavov , N . Strategies for Increasing the Depth and Throughput of Protein Analysis by plexDIA . en. Journal of Proteome Research 22 , 697 – 705 . ( 2023 ) ( Mar . 2023). OpenUrl CrossRef PubMed 37. ↵ Kuehnbaum , N. L. , Kormendi , A. & Britz-McKibbin , P . Multisegment Injection-Capillary Electrophoresis-Mass Spectrometry: A High-Throughput Platform for Metabolomics with High Data Fidelity . Analytical Chemistry 85 , 10664 – 10669 ( Nov . 2013 ). OpenUrl CrossRef 38. Kuehnbaum , N. L. , Gillen , J. B. , Gibala , M. J. & Britz-McKibbin , P . Personalized Metabolomics for Predicting Glucose Tolerance Changes in Sedentary Women After High-Intensity Interval Training . Scientific Reports 4 , 6166 ( Aug. 2014 ). OpenUrl CrossRef PubMed 39. ↵ Kuehnbaum , N. L. et al. Multiplexed separations for biomarker discovery in metabolomics: Elucidating adaptive responses to exercise training . ELECTROPHORESIS 36 , 2226 – 2236 ( 2015 ). OpenUrl CrossRef PubMed 40. ↵ McDonnell , K. , et al. JMod: Joint modeling of mass spectra for empowering multiplexed DIA proteomics . bioRxiv ( 2025 ). 41. ↵ Meek , J. L . Prediction of peptide retention times in high-pressure liquid chromatography on the basis of amino acid composition . Proceedings of the National Academy of Sciences 77 , 1632 – 1636 ( Mar . 1980 ). OpenUrl Abstract / FREE Full Text 42. Petritis , K. et al. Improved Peptide Elution Time Prediction for Reversed-Phase Liquid Chromatography-MS by Incorporating Peptide Sequence Information . Analytical Chemistry 78 , 5026 – 5039 ( July 2006 ). OpenUrl CrossRef PubMed 43. Krokhin , O. V. Sequence-Specific Retention Calculator. Algorithm for Peptide Retention Prediction in Ion-Pair RP-HPLC: Application to 300- and 100-Å Pore Size C18 Sorbents . Analytical Chemistry 78 , 7785 – 7795 ( Nov . 2006 ). OpenUrl CrossRef PubMed 44. Klammer , A. A. , Yi , X. , MacCoss , M. J. & Noble , W. S . Improving Tandem Mass Spectrum Identification Using Peptide Retention Time Prediction across Diverse Chromatography Conditions . Analytical Chemistry 79 , 6111 – 6118 ( Aug . 2007 ). OpenUrl CrossRef PubMed 45. Moruz , L. , Tomazela , D. & Käll , L. Training, Selection, and Robust Calibration of Retention Time Models for Targeted Proteomics . Journal of Proteome Research 9 , 5209 – 5216 ( Oct . 2010 ). OpenUrl CrossRef PubMed 46. Ma , C. et al. Improved Peptide Retention Time Prediction in Liquid Chromatography through Deep Learning . Analytical Chemistry 90 , 10881 – 10888 ( Sept . 2018 ). OpenUrl CrossRef 47. Bouwmeester , R. , Gabriels , R. , Hulstaert , N. , Martens , L. & Degroeve , S . DeepLC can predict retention times for peptides that carry as-yet unseen modifications . Nature Methods 18 , 1363 – 1369 ( Nov . 2021 ). OpenUrl CrossRef PubMed 48. ↵ Zeng , W.-F. et al. AlphaPeptDeep: a modular deep learning framework to predict peptide properties for proteomics . en. Nature Communications 13 , 7238 . ( 2025 ) ( Nov . 2022). OpenUrl 49. ↵ Wallmann , G. et al. AlphaDIA enables End-to-End Transfer Learning for Feature-Free Proteomics en. June 2024 . 50. ↵ Navarro , P. et al. A multicenter study benchmarks software tools for label-free proteome quantification . eng. Nature Biotechnology 34 , 1130 – 1136 ( Nov . 2016 ). OpenUrl CrossRef PubMed 51. ↵ Peckner , R. et al. Specter: linear deconvolution for targeted analysis of data-independent acquisition mass spectrometry proteomics . Nature Methods 15 , 371 – 378 ( May 2018 ). OpenUrl CrossRef PubMed 52. ↵ Leduc , A. , Huffman , R. G. , Cantlon , J. , Khan , S. & Slavov , N . Exploring functional protein covariation across single cells using nPOP . Genome Biology 23 , 261 . ( 2023 ) ( Dec . 2022). OpenUrl 53. ↵ Leduc , A. , Khoury , L. , Cantlon , J. , Khan , S. & Slavov , N . Massively parallel sample preparation for multiplexed single-cell proteomics using nPOP . Nature Protocols , 1 – 27 ( 2024 ). 54. ↵ Helmeczi , E. et al. A Software Tool for Rapid and Automated Preprocessing of Large-Scale Serum Metabolomic Data by Multisegment Injection-Capillary Electrophoresis-Mass Spectrometry . Analytical Chemistry 97 , 175 – 184 ( Jan . 2025 ). OpenUrl CrossRef 55. ↵ Leduc , A. , Xu , Y. , Shipkovenska , G. , Dou , Z. & Slavov , N . Limiting the impact of protein leakage in single-cell proteomics . Nature Communications 16 , 4169 ( 2025 ). OpenUrl CrossRef PubMed 56. ↵ Rauniyar , N. & Yates III , J. R . Isobaric labeling-based relative quantification in shotgun proteomics . Journal of proteome research 13 , 5293 – 5309 ( 2014 ). OpenUrl CrossRef PubMed 57. ↵ Specht , H. et al. Single-cell proteomic and transcriptomic analysis of macrophage heterogeneity using SCoPE2 . Genome Biology 22 ( 2021 ). 58. ↵ Dou , M. et al. Automated Nanoflow Two-Dimensional Reversed-Phase Liquid Chromatography System Enables In-Depth Proteome and Phosphoproteome Profiling of Nanoscale Samples . Analytical Chemistry 91 , 9707 – 9715 . ( 2024 ) ( Aug . 2019). OpenUrl 59. ↵ Ramachandran , P. , Zoph , B. & Le , Q. V. Searching for Activation Functions Oct . 2017 . 60. ↵ Kingma , D. P. & Ba , J. Adam: A Method for Stochastic Optimization Jan . 2017 . 61. ↵ Abdrakhimov , D. A. et al. Biosaur: An open-source Python software for liquid chromatography–mass spectrometry peptide feature detection with ion mobility support . Rapid Communications in Mass Spectrometry n/a , e9045 . 62. ↵ Specht , H. , et al. Automated sample preparation for high-throughput single-cell proteomics en. Aug . 2018. ( 2022 ). 63. ↵ Cox , J. et al. Accurate Proteome-wide Label-free Quantification by Delayed Normalization and Maximal Peptide Ratio Extraction , Termed MaxLFQ. Molecular & Cellular Proteomics: MCP 13 , 2513 – 2526 . ( 2022 ) ( Sept . 2014). OpenUrl 64. ↵ Johnson , W. E. , Li , C. & Rabinovic , A . Adjusting batch effects in microarray expression data using empirical Bayes methods . eng. Biostatistics (Oxford, England) 8 , 118 – 127 ( Jan . 2007 ). OpenUrl CrossRef PubMed Web of Science 65. ↵ Leek , J. T. , Johnson , W. E. , Parker , H. S. , Jaffe , A. E. & Storey , J. D . The sva package for removing batch effects and other unwanted variation in high-throughput experiments . Bioinformatics 28 , 882 – 883 . ( 2024 ) ( Mar . 2012). OpenUrl 66. ↵ Ammar , C. , Gruber , M. , Csaba , G. & Zimmer , R . MS-EmpiRe Utilizes Peptide-level Noise Distributions for Ultra-sensitive Detection of Differentially Expressed Proteins[S]. English . Molecular & Cellular Proteomics 18 , 1880 – 1892 . ( 2024 ) ( Sept . 2019). OpenUrl View the discussion thread. Back to top Previous Next Posted May 27, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Increasing mass spectrometry throughput using time-encoded sample multiplexing Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Increasing mass spectrometry throughput using time-encoded sample multiplexing Jason Derks , Kevin McDonnell , Nathan Wamsley , Peyton Stewart , Maddy Yeh , Harrison Specht , Nikolai Slavov bioRxiv 2025.05.22.655515; doi: https://doi.org/10.1101/2025.05.22.655515 Share This Article: Copy Citation Tools Increasing mass spectrometry throughput using time-encoded sample multiplexing Jason Derks , Kevin McDonnell , Nathan Wamsley , Peyton Stewart , Maddy Yeh , Harrison Specht , Nikolai Slavov bioRxiv 2025.05.22.655515; doi: https://doi.org/10.1101/2025.05.22.655515 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Bioengineering Subject Areas All Articles Animal Behavior and Cognition (7629) Biochemistry (17660) Bioengineering (13881) Bioinformatics (41911) Biophysics (21436) Cancer Biology (18578) Cell Biology (25482) Clinical Trials (138) Developmental Biology (13371) Ecology (19887) Epidemiology (2067) Evolutionary Biology (24302) Genetics (15599) Genomics (22482) Immunology (17728) Microbiology (40363) Molecular Biology (17163) Neuroscience (88536) Paleontology (666) Pathology (2830) Pharmacology and Toxicology (4821) Physiology (7637) Plant Biology (15129) Scientific Communication and Education (2045) Synthetic Biology (4290) Systems Biology (9817) Zoology (2269)