De novo design of light-regulated dynamic proteins using deep learning

De novo design of light-regulated dynamic proteins using deep learning | 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 De novo design of light-regulated dynamic proteins using deep learning L. Scutteri , L.A. Abriata , S. Zhang , View ORCID Profile A. Hacisuleyman , View ORCID Profile K. Lau , F. Pojer , View ORCID Profile S.J. Rahi , P. Barth doi: https://doi.org/10.1101/2025.08.12.669910 L. Scutteri 1 Institute of Bioengineering , EPFL, Switzerland 2 Institute of Physics , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site L.A. Abriata 1 Institute of Bioengineering , EPFL, Switzerland 3 Protein Production and Structure Core Facility , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site S. Zhang 1 Institute of Bioengineering , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site A. Hacisuleyman 1 Institute of Bioengineering , EPFL, Switzerland 4 Department of Computational Biology , UNIL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for A. Hacisuleyman K. Lau 3 Protein Production and Structure Core Facility , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for K. Lau F. Pojer 3 Protein Production and Structure Core Facility , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site S.J. Rahi 2 Institute of Physics , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for S.J. Rahi P. Barth 1 Institute of Bioengineering , EPFL, Switzerland Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: patrick.barth{at}epfl.ch Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Recent advances in deep learning have enabled accurate design of static protein structures, but the de novo design of protein functions controlled by programmable, intramolecular conformational changes remains an unsolved challenge. Here, we present a general deep learning-guided framework for designing dynamic, multi-domain proteins allosterically regulated by light. By integrating photoresponsive domains into de novo scaffolds, we engineered conformational switches that exhibit precise, reversible structural transitions upon illumination. Structural, spectroscopic, and functional analyses validated our designs and demonstrated precise spatio-temporal optogenetic control of diverse cellular processes, including subcellular localization, intercellular signaling, and population-level behaviors. This work establishes a broadly applicable strategy for encoding long-range allosteric control through designed intramolecular motions, and opens new avenues for programming dynamic protein functions from first principles, with implications for basic research, synthetic biology, and therapeutic development. Introduction Protein switches play essential roles in biology by coupling input signals to output responses through structural rearrangements. Conformational transitions between distinct protein states regulate a broad array of functions, including enzyme catalysis( 1 ), signal transduction( 2 , 3 ), energy conversion( 4 ), or protein recognition( 5 ). Owing to their versatility, natural protein switches have become attractive targets for de novo protein design. However, engineering synthetic proteins with analogous switch-like dynamics remains a major challenge as effective switches must access multiple conformational states, display fine-tuned energetics between states, and respond to specific external cues( 6 ). Previous advances include Ca²⁺-responsive proteins( 7 ), stimulus-responsive two-state hinge proteins( 8 ), fold switches( 9 ), tunable protein biosensors( 10 ), and allosterically switchable protein assemblies( 11 ). Yet, these examples primarily rely on rigid-body movements driven by intermolecular interactions or localized intramolecular rearrangements. In contrast, natural protein switches frequently integrate stimulus sensing and response at distant sites, enabling long-range signal transmission( 12 – 15 ), and the evolution of modular, multi-domain architectures( 16 – 18 ). To date, such fundamental allosteric mechanisms involving the long-range dynamic communication across protein structures have not been recapitulated by de novo design( 19 – 22 ). Biomolecular switches’ functions can be triggered and regulated by a variety of inputs, including ligand binding, environmental factors (e.g., pH, heat, force), allosteric mutations( 23 ), and post- translational modifications. However, these conventional stimuli often pose practical limitations: they can be slow, irreversible, labor-intensive, or invasive, which limits their applicability. In contrast, optogenetics leverage light as a precise, fast, reversible, tunable, non-invasive, and cost-effective stimulus for controlling protein behavior( 24 – 28 ). Genetically encoded light- sensitive domains—such as the LOV2 domain from Avena sativa phototropin 1 (AsLOV2)—can be inserted into target effectors to confer light-dependent regulation, enabling tight spatio- temporal control over protein function and cellular processes( 29 – 38 ). Despite advances in predicting insertion sites computationally( 39 ), reliably engineering efficient allosteric coupling with LOV2 remains difficult, often requiring extensive experimental screening or rounds of in vivo evolution and optimization( 40 – 42 ). To date, the rational, computational design of optogenetic switches with tunable, allosterically controlled conformational dynamics and function has not been achieved. Results De novo design approach of multi-domain dynamic protein optoswitches To address these limitations, we set out to develop a general computational framework for designing multi-domain dynamic protein optoswitches, in which a stimulus-sensing domain allosterically regulates the function of a responder region through long-range intramolecular conformational changes ( Fig. 1a ). Our design strategy was guided by three key criteria: 1) To enable switching between well-folded conformations, we focused on de novo design of globular responder scaffolds that remain stably folded even upon disruption of intramolecular interactions between switchable elements. 2) To achieve allosteric regulation of the responder conformation by the sensor domain, we designed scaffolds in which each switchable region is allosterically coupled to one end of the sensor domain. 3) To encode light-inducible protein function, we aimed to engineer conformational transitions that selectively expose or occlude a functional epitope. Download figure Open in new tab Fig. 1. Deep learning-guided de novo design of dynamic protein optoswitches. ( a ) Design concept for engineering multi-domain dynamic protein optoswitches that interconvert between two states, enabling allosteric control of a responder domain by a light-responsive sensor. ( b-d ) General framework for deep learning-guided de novo design of optoswitchable proteins. ( b ) De novo design of scaffolds that undergo controlled conformational switching between distinct states. ( c ) Allosteric coupling to enable optogenetic control through deep learning-guided linker design and Mutual Information (MI) analysis to select linkers that precisely modulate mechanical coupling between the optosensor and the switchable scaffold. Tuning of the coupling strength between domains allows fine adjustment of responder activity. ( e ) Experimental validation of optoswitch structure and dynamics by NMR spectroscopy and a split-sfCherry2 complementation assay. ( f ) Applications of programmable spatio-temporal optogenetic control of cellular behaviors through functionalized optoswitches. In the dark, a binding epitope grafted onto the switchable region is caged by the responder domain. Upon blue light illumination, the LOV2-mediated conformational change exposes the functional motif, eliciting the desired cellular response. Our approach involves three main steps ( Fig. 1b, c, d, e , Fig. S1 ): Step 1 focuses on the de novo design of the switchable responder scaffold ( Fig. 1b ). We begin by exploring sequence-structure space to generate scaffolds that stably fold into a single-state conformation in which the functional epitope is occluded (closed state). To convert these static scaffolds into dynamic, switchable systems, we computationally design an ensemble of alternative conformational states that expose the functional epitope (open state). This is achieved by combining steered molecular dynamics (MD) simulations with deep learning-based backbone generation. Switchable sequences capable of adopting both conformations are then identified through a combination of deep learning and physics-based sequence optimization methods. Step 2 centers on the assembly and design of the allosterically coupled sensor-responder pair ( Fig. 1c ). To ensure efficient control of responder conformations, the sensor domain is inserted between the switchable elements of the responder scaffold. The strength of allosteric regulation is governed by the structural coupling between the two domains, which can be tuned by modifying the properties of the interdomain linkers. Increased linker rigidity promotes stronger coupling and vice versa. These linkers are de novo designed using iterative cycles of deep learning–based backbone and sequence generation, followed by evaluation of interdomain coupling using Mutual Information (MI) analyses from physics-based simulations( 2 , 43 ). The allosteric coupling should ultimately dictate the level of light-induced control of the responder activity through shifts in the population of molecules occupying each conformational state ( Fig. 1d ). Step 3 involves comprehensive experimental characterization of the designed switches ( Fig. 1e ). Responder conformations and their intrinsic dynamics are analyzed using NMR spectroscopy and physics-based simulations. The fine-tuned energetics and population distribution of the responder and responder-sensor conformational states are validated through real-time fluorescence measurements in living cells. As a proof of concept, we designed light-sensitive switches using AsLOV2 as the stimulus- sensing domain. Upon blue light irradiation, AsLOV2 undergoes a conformational change that releases its C-terminal Jα helix from the core domain. Building on this mechanism, we engineered responder scaffolds in which the switchable region containing the functional epitope is allosterically coupled to the C-terminus of AsLOV2, enabling light-dependent functional control. Specifically, we designed a scaffold structure bearing a switchable helix that remains caged within the globular core of the responder in the dark but becomes solvent-exposed upon illumination. This architecture provides a versatile platform for the optical regulation of cellular interactions mediated by binding epitopes engineered onto the switchable helix. In principle, our approach allows the custom design of diverse scaffold topologies and optogenetic control systems, enabling precise spatio-temporal regulation of programmable cellular functions ( Fig. 1f ). Conformations and dynamics of the de novo responder switch We first examined whether a de novo designed responder domain with α/β topology could be sculpted to adopt distinct closed (state 1, Fig. 2a ) and open (state 2, Fig. 2b ) conformations. To this end, we selected sequences computationally predicted to favor either state, expressed them, and assessed their folding stability. CD spectra confirmed the presence of well-folded structures in both states, in agreement with the designed αβ topology ( Fig. S2b ). Thermal unfolding profiles indicated exceptional stability, with melting temperatures exceeding 100LJ°C for both closed and open states ( Fig. S2c ). Remarkably, the open state design retained proper folding at elevated temperatures, despite the exposed core of the protein and the dynamic nature of the C-terminal helix. Download figure Open in new tab Fig. 2. Structural analysis of de novo designed dynamic protein switches ( a-b ) Schematic energy landscapes of the de novo designed scaffolds in the closed and open states. ( c-d ) NMR ensembles of the de novo designed scaffolds in the closed and open states. Representative long-range NOE distance restraints for residues 115 and 123 of the switchable helix in the closed and open states. Long-range NOEs between the switchable helix and the β-sheet core are exclusively observed in the closed state scaffold. ( f ) 1 H, 15 N Heteronuclear NOE for the closed (blue) and open (red) states of the scaffold. Structural models are colored by per-residue Het NOE values, highlighting fast-timescale dynamics of the C- terminal helix in the open state. ( g ) Per-residue root mean square fluctuation (RMSF) profiles for the closed (blue) and open (red) states of the scaffold, averaged over four independent 500 ns molecular dynamics (MD) simulations. ( h ) Conformational energy landscape describing the equilibrium between closed and open states of the responder domain. Designed allosteric mutations at the helix-core interface allow precise fine-tuning of the scaffold energetics and conformational equilibrium, enabling the creation of switchable responder domains. ( i ) Experimental validation of the switchable responder conformational equilibrium through sfCherry2-based split complementation cell-based assay. In the closed state, the spatial proximity of the N- and C-termini of the scaffold enables reconstitution of the split sfCherry2, producing a fluorescent signal. As the conformational equilibrium shifts towards the open state, increasing fraction of scaffolds will uncage the switchable element preventing sfCherry2 reconstitution and leading to overall decreasing fluorescence signals. On the left Y-axis, sfCherry2-based signal of designed scaffolds in HEK293T cells, reflecting the distribution of states for each variant. On the right Y-axis, ΔE (in Rosetta Energy Units: reu) between closed and open states calculated using the Rosetta software. We then investigated the closed and open conformations of the responder domain using nuclear magnetic resonance (NMR) spectroscopy. We acquired 2D 1 H, 15 N-heteronuclear single-quantum coherence (HSQC) spectra for each design and 3D 1 H, 15 N, 13 C NMR experiments for the open state sequence ( Methods , Fig. S3 ). The NMR ensembles for both states matched closely the design models (Cα RMSD < 2.5Å). While the switchable element is well-folded onto the core of the scaffold in the closed state ( Fig. 2c ), it is largely disengaged in the open state ( Fig. 2d ). This difference stems from a large set of long-range NOEs connecting the switchable helix (residues 112-126) to the protein core in the closed state ( Fig. 2e , Fig. S4a ), and a lack of corresponding restraints in the open state design. The fully disengaged helix conformation observed for the open state is in excellent agreement with the design models. To directly probe the conformational dynamics of each state, we collected 15 N T 1 , 15 N T 2 and { 1 H, 15 N} heteronuclear NOE relaxation data. The resulting R 2 /R 1 profiles ( Fig. S4b ) displayed median values consistent with monomeric states for both proteins, along with reductions indicative of fast (ps–ns) motions at and around the C-terminal helix in the open state only. The heteronuclear NOE measurements ( Fig. 2f ) corroborated these findings, confirming that such ps–ns dynamics occur exclusively in the open state. This dynamic signature is consistent with the C-terminal helix being detached from the core, as both designed and observed in the NMR structure. In contrast, the closed-state design exhibited high rigidity at the switchable helix–core interface, in agreement with its tightly packed conformation. Notably, 15 N CEST experiments revealed no exchange processes on the slower μs–ms timescale, and the R₂/R₁ ratios showed no significant positive deviations, consistent with the absence of slow conformational dynamics in the open-state design. In line with the design goals, all metrics supported a well-folded and conformationally stable core of the scaffold in the open state despite the absence of stabilizing interactions with the switchable helical subdomain. Secondary structure analysis based on ΔCα–ΔCβ chemical shifts revealed excellent agreement with the expected topology in both conformations ( Fig. S4c ). Most importantly, the C-terminal helix retained significant helical propensity in the open state, consistent with its folded yet flexible behavior observed in the NMR ensemble. Lastly, dynamic changes between the states localized to the switchable helix were corroborated by root-mean-square fluctuation (RMSF) values obtained from molecular dynamics (MD) simulations carried out on each design model ( Fig. 2g ). These results provide direct structural evidence that our approach enables the de novo design of novel protein topologies with programmable switchable elements adopting distinct conformations with near-atomic precision. Fine-tuning conformational equilibrium of the responder by multistate design While the NMR experiments validated sequences that individually adopted either conformational state of the responder switch, we next aimed to identify sequences capable of toggling between both. We reasoned that searching for sequences spanning a broad range of conformational energy differences (ΔE, Fig. 2h ) would increase the likelihood of discovering such dynamic switches. To this end, we focused on residues at the interface between the protein core and the switchable helix in the closed state and performed multistate design calculations, combining classical Rosetta-based energy optimization with deep learning-guided sequence design, to identify mutations compatible with both conformational states. This strategy yielded 15 switchable sequence variants that span a gradient of state occupancies between fully closed and fully open conformations. These designs differ from the NMR-validated closed-state sequence by 3 to 23 mutations and incorporate polar substitutions that enhance solubility and stability in the open state, as well as van der Waals packing defects that destabilize the helix- core interface, thereby promoting conformational flexibility ( Methods ). To directly monitor the conformational switching of our de novo designs, we established a real- time, non-invasive cellular assay capable of efficiently quantifying state distributions across numerous variants. This assay is based on split complementation of the sfCherry2 fluorescent protein, whose two fragments (1–10 and 11) spontaneously associate into a functional fluorophore when brought into close proximity( 44 , 45 ). Fluorescence is observed in the closed conformation of the scaffold, in which the N- and C-termini are juxtaposed, allowing efficient complementation. In the open conformation, displacement of the 11 th β-strand disrupts complementation, leading to fluorescence loss ( Methods ). Because the assay measures the average fluorescence from all protein switches expressed in a cell, the split sfCherry2 readout serves as a sensitive and direct reporter of the protein’s conformational state distribution. We applied the sfCherry2 complementation assay to our de novo responder library. The NMR- validated sequence, which populates the closed state, exhibited the highest fluorescence, consistent with close N- and C-terminal proximity. In contrast, multistate designs showed progressively reduced fluorescence, correlating with decreasing ΔE values ( Fig. 2i , Fig. S5 ). These results confirm that multistate designs function as conformational switches and demonstrate that our approach can de novo design dynamic proteins with precise state distribution. Optogenetic control of de novo responder switching dynamics We next set out to engineer light-responsive switches by designing LOV2 sensor-responder fusion proteins ( Fig. 3a, b ). To demonstrate full programmable optical control over responder conformational distributions, we fused LOV2 to responders that intrinsically adopt either the closed or open state. Our goal was to achieve optimal allosteric coupling between the two domains such that LOV2 excitation would uncage the switchable element in the closed-state responder while maintaining its closed conformation in the dark ( Fig. 3a ). Conversely, a successful light-controlled switch based on the open-state responder would cage the switchable element in the dark while preserving the intrinsic open conformation under illumination ( Fig. 3b ). Download figure Open in new tab Fig. 3. Designing optimal light-regulated allosteric control of de novo responder switching dynamics ( a-b ) Schematic representation of the expected impact of LOV2 grafting onto closed ( a ) and open ( b ) state sfCherry2-responder domains. Expected fluorescence signals due to sfCherry2 complementation are displayed as a function of the allosteric coupling between the light sensor and responder domains. The stronger the coupling, the higher the light-induced shift in responder conformations and changes in sfCherry2 reconstitution. AlphaFold structural models of the sfCherry2-responder domains (left) and LOV2-responder fusions with strongest predicted allosteric coupling (right). The models are color-coded by domain separation (sfCherry2 β-barrel and 11 th β-strand, red; responder core, gray; LOV2, blue; switchable element, pink). ( c ) Deep learning-guided fine-tuning of allosteric coupling between the stimulus-sensing domain and the de novo responder scaffold, and optogenetic control of dynamic switching behavior from the closed state (dark) to the open state (light). ( d ) Correlation between allosteric coupling (inferred by Mutual Information (MI) between the stimulus-sensing domain and the switchable helix of the responder), and fluorescence ratio from dark to light for different linker designs (R 2 = 0.947, p = 0.0011). ( e ) Validation of the reversible light-induced switching behavior of the dynamic protein optoswitches derived from the closed and open states of the de novo designed scaffolds. Initial dark phase marked in blue, light-induction (10’) marked in magenta, dark-reversion (10’) marked in pink. ‘ns’ = not significant, *p-valueLJ<LJ0.05, **p-valueLJ<LJ0.005, ***p-valueLJ<LJ0.001, ****p- value < 0.0001. Direct fusion of LOV2 to the closed state responder, without intervening linkers, introduced flexible junctions between the two domains, which drastically reduced mechanical coupling and resulted in sub-optimal light responsiveness. To overcome this and achieve optimal light- inducible control, we computationally designed a set of de novo linker regions connecting the N- and C-termini of LOV2 to the core and switchable subdomains of the closed state responder scaffold. This involved cycles of deep learning-based structure prediction and sequence design to evaluate the structural impact of LOV2 grafting and eliminate linker designs that significantly perturbed the responder conformation in the dark. We generated a diverse library of optoswitches incorporating linkers of varying length and conformational flexibility to explore a wide range of allosteric coupling efficiencies. We then performed physics-based simulations on six representative fusions to compute Mutual Information (MI) between the LOV2 stimulus- sensing domain and the switchable region of the responder. These MI values served as a quantitative proxy for allosteric coupling strength between the two domains. To assess functional coupling, we fused split sfCherry2 to the responder domains and measured fluorescence changes in yeast upon blue light stimulation ( Methods , Fig. 3c ). The intrinsically low affinity between the sfCherry2 fragments allows reversible complementation and dissociation, without constraining the light-regulated conformational dynamics( 46 , 47 ). All designs exhibited significant light-dependent fluorescence shifts, indicating effective optogenetic control. Notably, MI values strongly correlated with fluorescence changes ( Fig. 3d ), validating our computational framework for achieving precise and programmable allosteric control through de novo designed linkers. Among the tested constructs, Linker #5, composed of a rigid, well- folded helical segment, showed the highest MI and was selected for further experimental validation. Since dynamic reversibility is a key requirement for optogenetic control, we measured the kinetics of the optoswitches under brief blue light exposure ( Fig. 3e ). Upon illumination, the LOV2-controlled optoswitches exhibited a complete cycle of fluorescence decrease and recovery within 10 minutes, confirming dynamic, reversible, and light-dependent modulation of scaffold conformation. We next assessed whether LOV2 fusion to the open state responder would enable light inducible switchable behavior. We engineered LOV2 grafting to allosterically position the switchable helix in close proximity to the responder β-core in the dark, while enabling restoration of an open conformation under light. Consistent with the design goals, we observed sfCherry2 complementation in the dark and a large reversible decrease of fluorescence signal under blue light illumination ( Fig. 3e ). While the strong fluorescence shifts confirmed light-induced conformational transitions of the switchable helix, we next examined how LOV2 fusion affected the responder’s structural integrity. The 1 H, 15 N-HSQC spectra of the Closed-LOV2 and Open-LOV2 fusion proteins displayed high chemical shift dispersion, consistent with well-folded structures ( Fig. S6, S7 ). To assess structural perturbations at the residue level, we computed minimal chemical shift perturbations (minCSPs) between the fusions and their isolated responder scaffolds. For both optoswitches, deviations were mainly confined to the β-core-helix interface, indicating overall fold preservation. In the Closed-LOV2 construct, moderate perturbations suggested a slight reorientation of the helix onto the β-core, as predicted by AlphaFold. By contrast, the Open- LOV2 design exhibited lower minCSPs localized to residues linking the core and switchable helix, consistent with occlusion of that region upon fusion to LOV2 but lack of strong interactions with the β-core. LOV2 grafting induced helix caging in both designs in the dark; however, the β-core-helix interface was stabilized by strong interactions only in the Closed-LOV2 switch, yielding exceptional thermal stability (Tm > 100 °C) and overall low conformational dynamics ( Fig. S8, Fig. S9 ). In the Open-LOV2 construct, the absence of stabilizing β-core-helix contacts led to selective unfolding of the sensor domain (Tm ≈ 50 °C, closed to its reported intrinsic stability( 48 , 49 )), while the responder core remained thermostable (Tm > 100 °C, Fig. S8 ). Simulations further showed high flexibility in the switchable helix yet persistent helical propensity, reflecting a folded but dynamic element largely decoupled from a rigid core ( Fig. S9 ). Together, these results demonstrate that our approach enables the de novo design of multi- domain protein switches that interconvert between defined conformations with near-atomic precision and allow finely tunable, optogenetic control of their conformational equilibria. Programmable spatio-temporal control of cellular behaviors via optoswitches functionalization The precise, dynamic, and reversible light-induced activation of our de novo switches offers a powerful strategy for programmable spatio-temporal control of cellular behaviors, enabled by the ability to design switchable scaffolds with custom shapes and epitope properties from first principles ( Fig. 1d ). As a proof of concept, we functionalized our optoswitches with photo-caged peptide binding motifs under the following design constraints: in the dark, the epitope must remain buried and tightly packed against the core of the de novo scaffold, rendering it inactive. Upon blue light exposure, the LOV2-driven conformational change should induce epitope exposure, thereby triggering a specific cellular response. This modular approach is broadly adaptable to diverse synthetic biology applications, establishing a versatile platform for engineering light-regulated cellular functions. To demonstrate the utility of our approach, we first applied it to the optogenetic control of subcellular localization ( Fig. 4a ). We grafted a nuclear localization signal (NLS)( 50 ) onto the C- terminal helix of the optoswitch scaffold, enabling light-inducible, reversible nuclear import with rapid kinetics on the minute timescale ( Fig. 4b ). In the dark, the NLS is sterically occluded, preventing its recognition by nuclear importins α and β. Upon blue light exposure, unfolding of the Jα helix unmasks the NLS, allowing engagement with the nuclear import machinery ( Fig. S10 ). This design yielded robust functionality, with a 5.6-fold increase in the nuclear-to- cytoplasmic (N:C) fluorescence ratio upon light stimulation, followed by a return to baseline in the dark ( Fig. 4b ). Additionally, nuclear import–export dynamics remained consistent over multiple illumination cycles, underscoring the robustness and reversibility of optogenetic subcellular localization control ( Fig. S11 ). We extended this strategy to the de novo Open-LOV2 scaffold functionalized with an NLS. Because of the high flexibility of its switchable helix, truncating four C-terminal residues was necessary to achieve effective caging in the dark. As expected, this design showed a higher baseline N:C ratio but an even stronger light-induced increase ( Fig. S12 ). Download figure Open in new tab Fig. 4. Functionalization of protein optoswitches for programmable spatio-temporal control of cellular behaviors. ( a ) Schematic of optogenetic control of sub-cellular localization. Photo-caging of the nuclear localization signal (NLS) enables reversible light-activated nuclear import of the dynamic protein optoswitches. ( b ) Time-lapse recording of light-induced nuclear import. Cells were incubated in the dark for 5LJmin before blue light irradiation for 15LJmin, followed by a 30 min dark-recovery phase. The fluorescent protein ymScarlet is fused at the N- terminus of the protein optoswitch. The relative nuclear-to-cytoplasmic (N:C) fluorescence ratio was quantified across frames of the time-lapse recording. ( c ) Schematic of optogenetic control of cell-cycle arrest. Reversible photo-caging of the α-pheromone peptide within the dynamic protein optoswitches enables light-activated arrest of the cell-cycle. ( d ) Time-lapse recording of optogenetic control of cell-cycle arrest. Cells were incubated in the dark for 10 h prior to galactose induction, followed by an additional 6LJh in the dark, and finally exposed to 8LJh of blue light. A secretion signal was fused to the N-terminus of the optoswitch to enable its extracellular export, allowing the light-dependent exposure of the C-terminal α-pheromone peptide to activate GPCR signaling at the cell surface. The relative number of cells was quantified across frames of the time-lapse recording. Together, these findings underscore the versatility of the designed optoswitch architecture. Depending on the desired cellular response, distinct combinations of optoswitch scaffolds and peptide cargos can be rationally engineered to favor either tighter caging in the dark or enhanced exposure upon illumination, enabling tunable optogenetic control over protein function. To further demonstrate the versatility of our optoswitch platform, we next targeted a fundamentally different cellular application: the optogenetic control of intercellular communication and population-level behavior. Specifically, we focused on the well- characterized mating pathway in Saccharomyces cerevisiae , in which exposure to α-pheromone secreted by MATα cells triggers G 1 cell-cycle arrest in MATa cells as part of the mating response( 51 , 52 ) ( Fig. 4c ). We engineered the system so that, in the dark, the α-pheromone epitope remains sterically occluded within the optoswitch scaffold, preventing its interaction with the Ste2 G protein- coupled receptor (GPCR)( 53 ). Upon blue light absorption and LOV2-mediated undocking of the Jα helix, the α-pheromone becomes solvent-exposed and capable of engaging Ste2, thereby initiating the downstream signaling cascade leading to cell-cycle arrest ( Fig. S13 ). To engineer a self-regulating cell population control system based on intercellular communication, we genetically programmed cells to express and secrete the optoswitch. To ensure proper secretion and surface-level activity, we appended a secretion signal peptide to the N-terminus of the optoswitch ( Methods ). Functionalization of the designed optoswitch with the α-pheromone thus enabled light-dependent control of cell-cycle progression at the G 1 phase in MATa cells. In the dark, cells proliferated normally ( Fig. S14 ); upon blue light illumination, however, we observed robust and reproducible cell-cycle arrest ( Fig. 4d ). Together, these results establish our de novo designed multi-domain optoswitches as a modular platform for rewiring diverse cellular responses with high spatio-temporal resolution. This approach enables precise, light-dependent control over intercellular signaling pathways and communication, offering a powerful framework for dynamic cellular engineering in synthetic biology. Discussion The field of de novo protein design has transformed our capacity to build synthetic proteins with defined structures and tailored functions. However, the rational design of dynamic, multi-state proteins capable of reversibly transitioning between distinct conformations has remained a longstanding challenge. Here, we introduce a general deep learning-guided framework for de novo designing dynamic, multi-domain, light-responsive protein switches. Our strategy integrates deep learning-based multistate sequence design and structure prediction with physics-based energy calculations and molecular dynamics simulations. Using this approach, we engineered several protein scaffolds that undergo precise and reversible conformational transitions on functional timescales upon light stimulation. By embedding functional binding epitopes into these scaffolds, we generated optoswitches that enable programmable, spatio- temporal control of cellular processes, including intracellular localization, intercellular signaling, and population-level dynamics. To our knowledge, this represents the first demonstration of optogenetically controlled de novo protein switches and their application in synthetic cell biology. Importantly, because scaffold shapes, binding surfaces, and conformational transitions can now be designed with atomic precision, our framework opens new frontiers for building light-controllable protein systems and dynamic cellular functions in synthetic biology ( Fig. 1d ). Our results revealed remarkable agreement across deep learning-based design predictions, physics-based simulations, experimentally determined structures, and observed conformational dynamics. Notably, we could rationally tune allosteric communication between sensor and responder domains, providing direct evidence that allosteric control can be designed de novo in multi-domain proteins. In contrast to prior efforts that confined conformational changes to localized structural elements, our designs exhibit long-range allosteric transitions spanning over 100 Å, comparable to the global intramolecular rearrangements seen in naturally evolved systems such as signaling receptors, motor proteins, and multi-domain enzymes. Because our approach is modular and agnostic to specific domain structure and function, it offers a generalizable strategy for engineering allosteric communication pathways across diverse multi-domain proteins. In principle, this framework can be applied to design any allosterically regulated protein switch integrating distinct sensor and responder modules, broadening its impact well beyond the field of optogenetics. Collectively, this work lays the foundation for programmable synthetic proteins with precisely tunable conformational dynamics, expanding the design space of synthetic biology, optogenetics, and protein engineering. Methods Generation and design of responder scaffold structures We first sought to identify a topology that enables: ( 1 ) conformational transitions between well- folded subregions, ( 2 ) insertion of a light-sensing module between these regions, and ( 3 ) functionalization of the switchable segment at its C terminus to confer light-dependent biological activity. A de novo designed αβ protein, featuring a C-terminal helix docked onto a β-core, fulfilled these design criteria and was selected as the closed-state scaffold for optoswitch construction( 54 ). We next aimed to construct physically-realistic backbone structures of the scaffold in an open state conformation where the C-terminal helix is unlocked from the β-core. To this end, we extracted the 20 conformers from the NMR ensemble of the de novo designed closed state scaffold. The structures were idealized and relaxed using Rosetta, and conformer 4 - exhibiting the lowest Rosetta energy - was selected as a representative closed state structure. To generate an open state from the closed state conformation, we employed steered molecular dynamics (MD) simulations (see Methods, molecular dynamics (MD) simulations), applying a pulling force along a reaction coordinate perpendicular to the C-terminal helix, while fixing the protein core as an anchor. Representative frames in which the C-terminal helix was displaced, yet the scaffold remained folded, were extracted from the simulation trajectory. To preserve the integrity of the core structure from any disruptions potentially introduced by the pulling force, we aligned the open and closed state structures, and grafted the C-terminal helix (residues 112- 130) from the open conformation back onto the closed state core structure (residues 1-109). The region linking the helix to the core (residues 110-111) was redesigned using RosettaRemodel( 55 ). To generate structural diversity around the open state conformation and ensure designability, we generated 10 structures using the partial diffusion mode of RFdiffusion( 56 ) and performed subsequent sequence design using ProteinMPNN( 57 ), yielding 1000 sequences in total which were refolded with ColabFold( 58 ) in single sequence mode and ranked based on the AlphaFold2( 59 ) confidence metric (pLDDT), and the top-scoring design representing the open state was selected for experimental characterization and for multistate design calculations. Multistate design of switchable scaffolds To generate sequences switching between closed and open states, we engineered mutations in the single state design to minimize the energy difference (ΔE) and enable reversible switching between the two states. We initially applied a Rosetta-based multistate design protocol, combining atomistic modeling with a genetic algorithm to optimize amino acid sequences across multiple conformations( 60 ). The following fitness function was employed, where M1 represents the initial energy of the model in the open state and M2 represents the initial energy of the model in the closed state (closed and open state structures were selected as described above): We also carried out multistate design using ProteinMPNN( 57 ), a deep learning sequence design method capable of directly generating designs compatible with distinct input states. To generate switchable sequences that modulate the conformational equilibrium specifically at the helix-core interface, we restricted the designable space to 24 residues in this region that encounter markedly different local biochemical environments due to conformational rearrangements between states. Residues were designed by jointly considering both states, using weighted averages of the ProteinMPNN logit distributions. To comprehensively explore and program the conformational equilibrium between states, we tested a broad range of state-weighting schemes: 100–0%, 75–25%, 50–50%, 25–75%, 10–90%, and 0–100%. For each weighting scheme, ProteinMPNN was used to generate 100 sequences, which were then subjected to further computational evaluation. First, we threaded the designed sequences back onto both closed and open state structures using Rosetta CoupledMoves, followed by model relaxation and energy evaluation to compute the energy difference (ΔE) between the two states. Representative energy values for each weight class are shown in Figure S5a . Second, we predicted the structures of the multistate designs using AlphaFold3, assessing switchability by biasing the input with either the closed or open state model as a template. Predicted structures, colored by pLDDT confidence scores, are shown in Figure S5b . Interestingly, only switches from the 10–90% weighting scheme exhibited switchable behavior dependent on provided templates according to AlphaFold3 predictions. Linker design for LOV2 sensor-responder fusion proteins Series of linker variants were designed to enhance signal transmission from the LOV2 Jα helix to the switchable element, while preserving the integrity of the scaffold’s closed conformation. The strength of allosteric regulation depends on the structural coupling between sensor and responder domains, which can be modulated by adjusting linker length, rigidity, and secondary structure. To optimize coupling, we explored multiple strategies by varying linker lengths and/or de novo designing residues within positions −2 to +2 at the sensor–responder junction. Sequence design was performed with ProteinMPNN, generating 1,000 variants that were ranked according to amino acid frequency distributions and the predicted helical propensity of the linker region computed using HHpred. Structural models of the LOV2 sensor–responder fusions were then predicted using AlphaFold3, and linker confidence (pLDDT) values were used to evaluate structural reliability. To estimate linker rigidity, we analyzed the structural similarity among the top five AlphaFold models as an indirect measure of conformational variability. Increased linker rigidity was inferred to promote stronger domain coupling, whereas higher flexibility indicated weaker coupling. Molecular dynamics (MD) simulations Simulations were performed using GROMACS 2020.5( 61 , 62 ) with the CHARMM36 all-atoms forcefield( 63 ). Systems were simulated in an NPT ensemble at 300LJK and 1LJbar, using a Nosé–Hoover thermostat for temperature control and a Parrinello–Rahman barostat with semi- isotropic pressure coupling (relaxation time: 5LJps). As for equilibrium Molecular dynamics (MD) simulations, following constrained equilibration, four or five independent production runs were performed for each system, each consisting of 500 ns trajectories. As for steered Molecular dynamics (MD) simulations, performed to generate open state conformations during the open state design process, a pulling force was applied along a reaction coordinate perpendicular to the C-terminal helix, using the protein core as a fixed anchor. Simulations were carried out with a pulling rate of 0.8 nm , and a spring constant of 1000 kJ , with each SMD trajectory lasting 10LJns. To quantify dynamic correlations between residues, we computed Mutual Information (MI) between time series of Cα atom distances. The trajectories were discretized into 15 bins, and MI was calculated for each residue pair using the scikit-learn Python package( 64 ). Molecular cloning The DNA sequence of the designs was synthesized by GenScript. DNA digestions were performed using restriction endonucleases (New England Biolabs, USA). PCRs were performed using high-fidelity Phusion Polymerase (New England Biolabs, USA). Cloning was performed using restriction enzyme digestion and ligation with T4 ligase (New England Biolabs) or Gibson assembly (New England Biolabs). The DNA sequences of the constructs were verified by Sanger sequencing (Microsynth AG, Switzerland). Protein expression and purification Plasmids were propagated in BL21(DE3) or NEB T7 Express lysY/Iq Escherichia coli cells cultured in LB medium, which was supplemented with kanamycin to select for plasmid- containing bacteria. Cells were cultured in Luria-Bertani (LB) medium with 50 μg/mL kanamycin at 37 °C overnight. The overnight culture was inoculated to 2-4 L of Terrific Broth (TB, Formedium) medium or M9 medium without ammonium chloride and glucose (Formedium), and then supplemented with 15N-ammonium chloride (1g/L) for 15N-labelled protein, or 15N- ammonium chloride (1g/L) and 13C6-Glucose (4g/L) for 15N-13C-labelled protein, with 50 μg/mL kanamycin and cultivated at 37 °C and 200 rpm until an optical density at 600 nm (OD600) of ∼0.8-1.0 was reached. Recombinant protein expression was induced upon addition of 0.5 mM isopropyl-β-D-thio-galactopyranoside (IPTG) and the culture was further incubated at 18 °C and 200 rpm for 18 h. The cells were harvested by centrifugation at 4,000 g for 20 min at 4 °C. The cell pellets were stored at -80 °C until further purification. For purification, the harvested cell pellet was resuspended in a lysis buffer (700 mM NaCl, 20 mM HEPES, pH 7.5, protease inhibitor tablet), and lysed by sonification (2.5 min, cycle 10s, 70% amplitude). The lysate was centrifuged at 20,000 g for 20 min at 4 °C. 25 mM Imidazole was added to the supernatant, and mixed with 5 mL HisPur Ni-NTA Resin beads (Thermo Scientific) or 5 mL of Ni-Indigo resin (Cube Biotech; 25 mM imidazole was omitted in this instance) for 2 hours at 4 °C. The supernatant was loaded onto a column, washed with increasing concentrations of imidazole and eluted in elution buffer (700 mM NaCl, 250 - 500 mM imidazole, 20 mM HEPES, pH 7.5). The eluted protein solution was buffer-exchanged and further purified by Size Exclusion Chromatography using a Superdex 200 Increase 10-300 GL column in 111 mM NaCl, 14.8 mM HEPES 7.5. Protein was concentrated to the appropriate concentrations for NMR studies and flash frozen in liquid N2 and finally stored at -80 °C. Yeast strains and media All experiments were conducted using the wild-type haploid W303 yeast strain background ( MATa ade2-1 leu2-3 ura3-1 trp1-1 his3-11,15 can1-100 ). Yeast transformations were performed following the LiAc/DNA carrier/PEG (polyethylene glycol) protocol( 65 ). For genomic integration, 1LJµg of linearized plasmid DNA was used to achieve single-copy insertion at the non-essential YCR051W locus. Cells were grown in synthetic complete (SC) medium lacking methionine (-Met), supplemented with 2% (w/v) glucose for transformation and selection purposes( 66 ). For preparing solid SC plates, 2% agar was added to the medium. Transformed strains were selected on drop-out plates lacking uracil. To induce expression of the protein of interest, which was placed under the control of the inducible GAL1 promoter, strains were cultured in SC -Met medium supplemented with 2% (w/v) raffinose. Protein expression was induced by the addition of 2% (w/v) galactose. Concerning the pheromone-induced cell-cycle control application, genomic integration of the construct under the control of the GAL1 promoter failed to produce sufficient arrest, likely due to low expression levels and extracellular dilution of the secreted optoswitch. To overcome this limitation, we used a high-copy 2μ plasmid to drive strong overexpression, which restored light- dependent arrest. In addition, deletion of BAR1 ( 67 , 68 ), an aspartyl protease responsible for α- pheromone degradation, was necessary to prevent loss of the signal and enable sustained pathway activation. HEK293T cell culture HEK293T cells were cultured in DMEM (Thermo Fisher Scientific) supplemented with 10% FBS (Thermo Fisher Scientific), 50 U/ml Penicillin-Streptomycin (Thermo Fisher Scientific) at 37LJ°C with 5% CO 2 . Cells were passaged every 2 or 3LJdays with 0.05% trypsin/EDTA. Lipofectamine 2000 (Thermo Fisher Scientific) was used to transiently transfect the HEK293T cells with 100 ng plasmid DNA. sfCherry2-based split complementation assay To monitor conformational changes of the designed protein optoswitches, we employed a split complementation assay based on the engineered sfCherry2 1–10/11 fluorescent protein, whose two fragments spontaneously associate to re-form a functional fluorescent protein when brought into close spatial proximity( 44 , 45 ). Importantly, sfCherry2 1-10/11 was specifically selected because the affinity between the two fragments is relatively low, leading to a sub-optimal complementation efficiency( 46 , 47 ). As a consequence, the interaction between the sfCherry2 subunits is not expected to significantly alter the conformational equilibrium between the closed and open states of the responder domain, and complementation efficiency should mostly reflect the spatial proximity of the switchable element relative to the scaffold core. In this system, the N-terminal β-barrel of sfCherry2 (sfCherry2 1-10 , 208 amino acids) is genetically fused to the N- terminus of the scaffold, while the eleventh β-strand of sfCherry2 (sfCherry2 11 , 16 amino acids) is genetically fused to the C-terminus of the scaffold. sfCherry2 complementation and resulting fluorescence signal reached its maximum in the closed conformation of the scaffold, where the N- and C-termini are juxtaposed. For switchable designs, the assay is expected to sensitively report conformational state distributions among the ensemble of proteins expressed at steady state in the cell, reflecting the Boltzmann distribution of states for each design. Constructs were expressed in HEK293T cells and in Saccharomyces cerevisiae . Fluorescence in HEK293 cells was quantified using a FlexStation 3 multi-mode microplate reader (Molecular Devices) with excitation at 560LJnm and emission at 610LJnm. Yeast fluorescence was analyzed by microscopy using a Nikon Ti2-E inverted fluorescence microscope equipped with appropriate filters for sfCherry2 detection. A red fluorescent protein reporter was specifically chosen to avoid spectral overlap with the excitation wavelength of the LOV2 domain (450 nm), ensuring compatibility with optogenetic applications. Fluorescence values were normalized by subtracting the background signal from the negative control, consisting of mock-transfected (HEK293T) or mock-transformed (yeast) cells carrying the empty vector. Circular Dichroism (CD) spectroscopy Purified protein variants were diluted to a final concentration of 10LJµM in buffer containing 10LJmM HEPES and 10LJmM NaCl at pHLJ7.5. Circular dichroism (CD) spectra were acquired using a Chirascan spectropolarimeter (Applied Photophysics) to monitor thermal denaturation. Measurements were performed from 20 °C to 98 °C in 2 °C increments, recording spectra between 200LJnm and 260LJnm. Solution NMR spectroscopy All NMR experiments were carried out in a Bruker spectrometer operating at 800.13 MHz 1 H frequency and equipped with a CPTC 1 H, 13 C, 15 N 5 mm cryoprobe controlled via an Avance Neo console. All experiments were carried out at 298 K on min. 800 µM protein solutions prepared in 111 mM NaCl, 14.8 mM HEPES at pH 7.5, which is the exact buffer in which the Closed state’s NMR assignments and structure were reported (PDB ID 2MR6, BMRB ID 25062). All spectra were acquired and processed with Bruker TopSpin 4 using standard pulse sequences, and analyzed with CARA, NMRtist/Cyana and Sparky-NMRFAM. Protein resonance assignment and structure determination on the open state scaffold Experiments for backbone resonance assignment consisted in standard triple resonance spectra HNCA, HN(CO)CA, HNCO, HN(CO)CA, CBCA(CO)NH and HNCACB acquired on a ca. 1 mM sample prepared from cultures grown in M9 medium supplemented with 13 C-enriched glucose and 15 N-enriched ammonium chloride. HCCH-TOCSY and 13 C-resolved NOESY experiments were recorded on the same sample used for backbone assignments; they served for side chain assignment and structure determination together with HNHA, 15 N-resolved NOESY and 15 N-resolved TOCSY spectra acquired on a 15 N-labeled sample plus 2D NOESY and 2D TOCSY spectra collected on an unlabelled sample. Spectra for backbone assignments were acquired with 40 increments in the 15 N dimension and 128 increments in the 13 C dimension, and processed with 128 and 256 points using forward linear prediction. The HCCH- TOCSY spectrum was recorded with 128 increments in the 13 C dimensions and processed with 256 and 512 points. 15 N-resolved NOESY and TOCSY spectra were acquired with 40 increments in 15 N and 128 increments in the indirect 1 H dimension, and processed with 128 and 256 points respectively. 1 H- 1 H 2D-NOESY and 2D TOCSY spectra were acquired with 512 increments in the indirect dimension, processed with 2048 points. Mixing times for NOESY spectra were 120 or 200 ms, and TOCSY spin lock times were 60 ms. All spectra were acquired and processed with Bruker’s TopSpin 4.0 using the standard pulse programs with sensitivity improvement. Backbone assignments were obtained in parallel through manual analysis with the program CARA( 69 ) and automatically with NMRtist( 70 ), a webserver implementation of the ARTINA( 71 ) pipeline that uses a deep neural network to identify signals in NMR spectra, FLYA( 72 ) for resonance assignment, and automatic NOE identification and analysis for structure determination using CYANA( 73 ). Both procedures converged to 90% of the 1 H, 15 N assignments for backbone units; the 14 residues without full backbone 1 H, 15 N assignments being residues 1 and 2, most of the C-terminal his-tag, the only proline, and the stretch Arg 117 -Arg 124 which covers part of the helix engineered to detach from the protein core. This stretch consists of the sequence REREEERR and is itself surrounded by other E and R residues, a complexity that prevented full assignments not only because of the repetitive nature itself but also because 13 CA and 13 CB shifts for Arg and Glu are quite similar. However, the central region of the HSQC spectrum displays a set of crosspeaks whose 13 C projections in the CA and CB axes of HNCA, HNCOCA, CBCACONH and HNCACB spectra are consistent with multiple consecutive pairs of residues that have similar CA/CA-1 (55.04 ppm) and similar CB/CB-1 (27.245 ppm), all around the values expected for glutamates and arginines and hence very likely belonging to residues of the Arg 117 -Arg 124 stretch. While we cannot assign these spin systems to specific residues of the stretch because of the excessive overlap, we can safely assume that all the CA and CB shifts recorded apply to all 8 residues of the region. These shifts were utilized to derive the ΔCα-ΔCβ values for this region shown in Figure S4c . Given NMRtist’s accuracy on backbone assignment on our spectra, fully matching our manual assignment, we then ran it on the full set of spectra to obtain automated side chain assignments and structure determination. NMRtist’s structure calculation procedure converged in the program’s two independent runs to the same structure, indicative of sufficient data quality even in the regions with partial assignments or with low numbers of NOE restraints due to dynamics. We took NMRtist’s Proposal 1 as the final structure, an ensemble calculated from 2060 NOE restraints and 242 dihedral angle restraints. More statistics related to structure determination are provided in Table S1 . 15 N relaxation on the closed and open states of the scaffold Heteronuclear { 1 H}- 15 N NOE was measured with 128 15 N increments processed with 256 points, using 16 scans, a saturation time of 6 seconds and a total recycle time of 10 seconds. 15 N T 1 and T 2 were measured with 128 15 N increments processed with 256 points and a total recycle time of 10 seconds, with the following delays: 8, 64 (twice), 136, 232, 336, 472, 664 (twice), 800, 1200 and 1600 ms for T 1 ; and 16, 32 (twice), 64, 96, 128, 160, 192 (twice), 224, 256 and 288 ms for T 2 . Spectra were acquired and processed in Bruker TopSpin and then analyzed with the relaxation module in Sparky-NMRFAM. Calculation of minimal chemical shift perturbations (minCSP) For each residue of the de novo designed responder domain in the closed or open conformation with assigned 1 H and 15 N resonances, we searched for the nearest crosspeak in the HSQC spectrum of the corresponding Closed-LOV2 or Open-LOV2 optoswitch, respectively, as measured by the regular Fluorescence microscopy Images were acquired using a Nikon Ti2-E inverted microscope (Nikon, Switzerland) equipped with a 60x oil, N.A. 1.40 objective and a Hamamatsu Orca-Flash 4.0 camera (Hamamatsu, Japan). The system was controlled via NIS-Elements software and maintained in focus using the Nikon Perfect Focus System. For time-lapse imaging and induction with diascopic light, a green band-pass interference filter (Nikon, Switzerland) was placed between the light source and the sample to prevent unintentional activation of the LOV2 optogenetic domain. During induction, diascopic light intensity was set to 80% of maximum. Acknowledgments We thank members of the Barth lab and the Rahi lab for helpful discussions. Funding : This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 945363; an EPFL Science Seed Fund awarded to P.B. and S.J.R.; P.B. is supported by Swiss National Science Foundation grants (31003A_182263 and 310030_208179), a Novartis Foundation for medical- biological Research grant 21C195, a Swiss Cancer Research grant KFS-4687-02-2019, funds from EPFL, and the Ludwig Institute for Cancer Research; SNSF grants CRSK-3_190526, 310030_204938, and CRSK-3_221036 awarded to S.J.R. Authors contributions: L.S., A.H. and P.B. designed the project; L.S. and S.Z. developed the computational framework; L.S. designed and screened dynamic protein switches, ran MD simulations and Mutual Information (MI) analysis, performed experimental validation of the designs. L.S. developed applications on programmable spatio-temporal controls of cellular behaviors under the supervision of S.J.R.; L.A. performed NMR spectroscopy analysis; K.L. and F.P. performed protein purification; L.S. performed biophysical characterization, aided by K.L. and F.P.; all authors participated in the analysis and interpretation of the results; L.S. and P.B. wrote the manuscript with contributions from the other authors; P.B. supervised the entire project. Competing interests: P.B. holds patents and provisional patent applications in the field of engineered T cell therapies and protein design. Data and materials availability: all data are available in the manuscript or the supplementary materials. Footnotes The structural and conformational dynamics characterization of the designed switches has been further extended. References 1. ↵ M. Huse , J. Kuriyan , The Conformational Plasticity of Protein Kinases . Cell 109 , 275 – 282 ( 2002 ). OpenUrl CrossRef PubMed Web of Science 2. ↵ R. E. Jefferson , A. Oggier , A. Füglistaler , N. Camviel , M. Hijazi , A. R. Villarreal , C. Arber , P. Barth , Computational design of dynamic receptor—peptide signaling complexes applied to chemotaxis . Nature Communications 2023 14 : 1 14, 1 – 17 ( 2023 ). OpenUrl PubMed 3. ↵ K.-Y. M. Chen , J. K. Lai , L. S. P. Rudden , J. Wang , A. M. Russell , K. Conners , M. E. Rutter , B. Condon , F. Tung , L. Kodandapani , B. Chau , X. Zhao , J. Benach , K. Baker , E. J. Hembre , & P. Barth , -Y M Chen , Computational design of highly signalling-active membrane receptors through solvent-mediated allosteric networks . Nature Chemistry 2025 17 : 3 17, 429 – 438 ( 2025 ). OpenUrl PubMed 4. ↵ M. Schliwa , G. Woehlke , Molecular motors . Nature 2003 422:6933 422, 759 – 765 ( 2003 ). 5. ↵ L. Gonzalez , J. J. Plecs , T. Alber , An engineered allosteric switch in leucine-zipper oligomerization . Nature Structural Biology 1996 3 : 6 3, 510 – 515 ( 1996 ). OpenUrl CrossRef PubMed Web of Science 6. ↵ R. G. Alberstein , A. B. Guo , T. Kortemme , Design principles of protein switches . Curr Opin Struct Biol 72 , 71 – 78 ( 2022 ). OpenUrl CrossRef PubMed 7. ↵ A. B. Guo , D. Akpinaroglu , C. A. Stephens , M. Grabe , C. A. Smith , M. J. S. Kelly , T. Kortemme , Deep learning-guided design of dynamic proteins . Science 388 , eadr7094 ( 2025 ). 8. ↵ F. Praetorius , P. J. Y. Leung , M. H. Tessmer , A. Broerman , C. Demakis , A. F. Dishman , A. Pillai , A. Idris , D. Juergens , J. Dauparas , X. Li , P. M. Levine , M. Lamb , R. K. Ballard , S. R. Gerben , H. Nguyen , A. Kang , B. Sankaran , A. K . Bera , 9. ↵ B. F. Volkman , J. Nivala , S. Stoll , D. Baker , Design of stimulus-responsive two- state hinge proteins . Science ( 1979 ) 381 , 754 – 760 ( 2023 ). OpenUrl CrossRef PubMed 9. ↵ B. Ruan , Y. He , Y. Chen , E. J. Choi , Y. Chen , D. Motabar , T. Solomon , R. Simmerman , T. Kauffman , D. T. Gallagher , J. Orban , P. N. Bryan , Design and characterization of a protein fold switching network . Nature Communications 2023 14 : 1 14, 1 – 14 ( 2023 ). OpenUrl PubMed 10. ↵ A. Quijano-Rubio , H. W. Yeh , J. Park , H. Lee , R. A. Langan , S. E. Boyken , M. J. Lajoie , L. Cao , C. M. Chow , M. C. Miranda , J. Wi , H. J. Hong , L. Stewart , B. H. Oh , D. Baker , De novo design of modular and tunable protein biosensors . Nature 2021 591 :7850 591, 482 – 487 ( 2021 ). OpenUrl CrossRef PubMed 11. ↵ A. Pillai , A. Idris , A. Philomin , C. Weidle , R. Skotheim , P. J. Y. Leung , A. Broerman , C. Demakis , A. J. Borst , F. Praetorius , D. Baker , De novo design of allosterically switchable protein assemblies . Nature 2024 632 :8026 632, 911 – 920 ( 2024 ). OpenUrl CrossRef PubMed 12. D. M. Rosenbaum , S. G. F. Rasmussen , B. K. Kobilka , The structure and function of G-protein-coupled receptors . Nature 2009 459 :7245 459, 356 – 363 ( 2009 ). OpenUrl CrossRef PubMed Web of Science 13. M. Zhang , T. Chen , X. Lu , X. Lan , Z. Chen , S. Lu , G protein-coupled receptors (GPCRs): advances in structures, mechanisms and drug discovery . Signal Transduction and Targeted Therapy 2024 9 : 1 9, 1 – 43 ( 2024 ). OpenUrl CrossRef 14. ↵ D. Wootten , A. Christopoulos , P. M. Sexton , Emerging paradigms in GPCR allostery: implications for drug discovery . Nature Reviews Drug Discovery 2013 12 : 8 12, 630 – 644 ( 2013 ). OpenUrl CrossRef PubMed 15. ↵ K. Y. M. Chen , J. K. Lai , L. S. P. Rudden , J. Wang , A. M. Russell , K. Conners , M. E. Rutter , B. Condon , F. Tung , L. Kodandapani , B. Chau , X. Zhao , J. Benach , K. Baker , E. J. Hembre , P. Barth , Computational design of highly signalling-active membrane receptors through solvent-mediated allosteric networks . Nature Chemistry 2025 17 : 3 17, 429 – 438 ( 2025 ). OpenUrl PubMed 16. S. J. Parsons , J. T. Parsons , Src family kinases, key regulators of signal transduction . Oncogene 2004 23 : 48 23, 7906 – 7909 ( 2004 ). OpenUrl CrossRef PubMed Web of Science 17. ↵ E. A. Tannous , L. A. Yates , X. Zhang , P. M. Burgers , Mechanism of auto-inhibition and activation of Mec1ATR checkpoint kinase . Nature Structural & Molecular Biology 2020 28 : 1 28, 50 – 61 ( 2020 ). OpenUrl PubMed 18. ↵ J. S. Lott , T. V. Lee , Revealing the Inter-Module Interactions of Multi-Modular Nonribosomal Peptide Synthetases . Structure 25 , 693 – 695 ( 2017 ). OpenUrl 19. J. Monod , J. P. Changeux , F. Jacob , Allosteric proteins and cellular control systems . J Mol Biol 6 , 306 – 329 ( 1963 ). OpenUrl CrossRef PubMed Web of Science 20. G. M. Süel , S. W. Lockless , M. A. Wall , R. Ranganathan , Evolutionarily conserved networks of residues mediate allosteric communication in proteins . Nature Structural Biology 2002 10 : 1 10, 59 – 69 ( 2002 ). OpenUrl CrossRef 21. ↵ N. Popovych , S. Sun , R. H. Ebright , C. G. Kalodimos , Dynamically driven protein allostery . Nature Structural & Molecular Biology 2006 13 : 9 13, 831 – 838 ( 2006 ). OpenUrl PubMed 22. ↵ J. F. Swain , L. M. Gierasch , The changing landscape of protein allostery . Curr Opin Struct Biol 16 , 102 – 108 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 23. ↵ D. Keri , LJ; Hijazi, $LJ; Oggier, ; Barth, § Affiliations, P. Barth , Computational rewiring of allosteric pathways reprograms GPCR selective responses to ligands . bioRxiv, 2022.03.30.486353 ( 2022 ). 24. J. Mathony , D. Niopek , Enlightening Allostery: Designing Switchable Proteins by Photoreceptor Fusion . Adv Biol 5 , 2000181 ( 2021 ). 25. A. C. McCue , B. Kuhlman , Design and engineering of light-sensitive protein switches . Curr Opin Struct Biol 74 , 102377 ( 2022 ). 26. V. Emiliani , E. Entcheva , R. Hedrich , P. Hegemann , K. R. Konrad , C. Lüscher , M. Mahn , Z. H. Pan , R. R. Sims , J. Vierock , O. Yizhar , Optogenetics for light control of biological systems . Nature Reviews Methods Primers 2022 2 : 1 2, 1 – 25 ( 2022 ). OpenUrl 27. ↵ V. Emiliani , E. Entcheva , R. Hedrich , P. Hegemann , K. R. Konrad , C. Lüscher , M. Mahn , Z. H. Pan , R. R. Sims , J. Vierock , O. Yizhar , Optogenetics for light control of biological systems . Nature Reviews Methods Primers 2022 2 : 1 2, 1 – 25 ( 2022 ). OpenUrl 28. ↵ V. Gligorovski , M. Labagnara , L. Scutteri , M. Blackholm , A. Möglich , N. Mansouri , S. J. Rahi , Light-directed evolution of dynamic, multi-state, and computational protein functionalities . bioRxiv , 2024.02.28.582517 ( 2025 ). 29. D. Niopek , D. Benzinger , J. Roensch , T. Draebing , P. Wehler , R. Eils , B. Di Ventura , Engineering light-inducible nuclear localization signals for precise spatiotemporal control of protein dynamics in living cells . Nature Communications 2014 5 : 1 5, 1 – 11 ( 2014 ). OpenUrl 30. D. Niopek , P. Wehler , J. Roensch , R. Eils , B. Di Ventura , Optogenetic control of nuclear protein export . Nature Communications 2016 7 : 1 7, 1 – 9 ( 2016 ). OpenUrl 31. K. L. Kroll , A. R. French , T. R. Sosnick , R. S. Rock , LILAC: enhanced actin imaging with an optogenetic Lifeact . Nature Methods 2023 20 : 2 20, 214 – 217 ( 2023 ). OpenUrl PubMed 32. M. Garcia-Marcos , K. Parag-Sharma , A. Marivin , M. Maziarz , A. Luebbers , L. T. Nguyen , Optogenetic activation of heterotrimeric g-proteins by LOV2GIVe, a rationally engineered modular protein . Elife 9 , 1 – 21 ( 2020 ). OpenUrl CrossRef PubMed 33. D. Strickland , Y. Lin , E. Wagner , C. M. Hope , J. Zayner , C. Antoniou , T. R. Sosnick , E. L. Weiss , M. Glotzer , TULIPs: tunable, light-controlled interacting protein tags for cell biology . Nature Methods 2012 9 : 4 9, 379 – 384 ( 2012 ). OpenUrl PubMed 34. S. Y. Lee , J. S. Cheah , B. Zhao , C. Xu , H. Roh , C. K. Kim , K. F. Cho , N. D. Udeshi , S. A. Carr , A. Y. Ting , Engineered allostery in light-regulated LOV-Turbo enables precise spatiotemporal control of proximity labeling in living cells . Nature Methods 2023 20 : 6 20, 908 – 917 ( 2023 ). OpenUrl PubMed 35. C. Renicke , D. Schuster , S. Usherenko , L. O. Essen , C. Taxis , A LOV2 Domain- Based Optogenetic Tool to Control Protein Degradation and Cellular Function . Chem Biol 20 , 619 – 626 ( 2013 ). OpenUrl CrossRef PubMed 36. F. Bubeck , M. D. Hoffmann , Z. Harteveld , S. Aschenbrenner , A. Bietz , M. C. Waldhauer , K. Börner , J. Fakhiri , C. Schmelas , L. Dietz , D. Grimm , B. E. Correia , R. Eils , D. Niopek , Engineered anti-CRISPR proteins for optogenetic control of CRISPR–Cas9 . Nature Methods 2018 15 : 11 15, 924 – 927 ( 2018 ). OpenUrl PubMed 37. ↵ H. Yumerefendi , A. M. Lerner , S. P. Zimmerman , K. Hahn , J. E. Bear , B. D. Strahl , B. Kuhlman , Light-induced nuclear export reveals rapid dynamics of epigenetic modifications . Nature Chemical Biology 2016 12 : 6 12, 399 – 401 ( 2016 ). OpenUrl PubMed 38. ↵ O. I. Lungu , R. A. Hallett , E. J. Choi , M. J. Aiken , K. M. Hahn , B. Kuhlman , Designing Photoswitchable Peptides Using the AsLOV2 Domain . Chem Biol 19 , 507 – 517 ( 2012 ). OpenUrl CrossRef PubMed 39. ↵ B. Wolf , P. Shehu , L. Brenker , A. L. von Bachmann , A. S. Kroell , N. Southern , S. Holderbach , J. Eigenmann , S. Aschenbrenner , J. Mathony , D. Niopek , Rational engineering of allosteric protein switches by in silico prediction of domain insertion sites . Nature Methods 2025 22 : 8 22, 1698 – 1706 ( 2025 ). OpenUrl PubMed 40. N. T. Southern , A. von Bachmann , A. Hovsepyan , M.-L. Griebl , B. Wolf , N. Lemmen , A.-S. Kroell , S. Westermann , J. Mathony , D. Niopek, Phage-Assisted Evolution of Allosteric Protein Switches . bioRxiv, 2025.06.12.659331 ( 2025 ). 41. ↵ E. Peter , B. Dick , S. A. Baeurle , Mechanism of signal transduction of the LOV2-Jα photosensor from Avena sativa . Nature Communications 2010 1 : 1 1, 1 – 7 ( 2010 ). OpenUrl PubMed 42. ↵ C. Kim , S. R. Yun , S. J. Lee , S. O. Kim , H. Lee , J. Choi , J. G. Kim , T. W. Kim , S. You , I. Kosheleva , T. Noh , J. Baek , H. Ihee , Structural dynamics of protein-protein association involved in the light-induced transition of Avena sativa LOV2 protein . Nature Communications 2024 15 : 1 15, 1 – 15 ( 2024 ). OpenUrl PubMed 43. ↵ K. Y. M. Chen , D. Keri , P. Barth , Computational design of G Protein-Coupled Receptor allosteric signal transductions . Nature Chemical Biology 2019 16 : 1 16, 77 – 86 ( 2019 ). OpenUrl PubMed 44. ↵ S. Feng , S. Sekine , V. Pessino , H. Li , M. D. Leonetti , B. Huang , Improved split fluorescent proteins for endogenous protein labeling . Nature Communications 2017 8 : 1 8, 1 – 11 ( 2017 ). OpenUrl PubMed 45. ↵ S. Feng , A. Varshney , D. Coto Villa , C. Modavi , J. Kohler , F. Farah , S. Zhou , N. Ali , J. D. Müller , M. K. Van Hoven , B. Huang , Bright split red fluorescent proteins for the visualization of endogenous proteins and synapses . Communications Biology 2019 2 : 1 2, 1 – 12 ( 2019 ). OpenUrl PubMed 46. ↵ S. Zhou , S. Feng , D. Brown , B. Huang , Improved yellow-green split fluorescent proteins for protein labeling and signal amplification . PLoS One 15 , e0242592 ( 2020 ). OpenUrl CrossRef PubMed 47. ↵ R. Kamiyama , K. Banzai , P. Liu , A. Marar , R. Tamura , F. Jiang , M. A. Fitch , J. Xie , D. Kamiyama , Cell-type-specific, multicolor labeling of endogenous proteins with split fluorescent protein tags in Drosophila . Proc Natl Acad Sci U S A 118 , e2024690118 ( 2021 ). OpenUrl Abstract / FREE Full Text 48. ↵ K. Felčíková , A. Hovan , M. Polák , D. S. Loginov , V. Holotová , C. Díaz , T. Kožár , O. S. Lee , R. Varhač , P. Novák , G. Bánó , E. Sedlák , Design of AsLOV2 domain as a carrier of light-induced dissociable FMN photosensitizer . Protein Science 33 , e4921 ( 2024 ). OpenUrl PubMed 49. ↵ S. Hemmer , N. E. Siedhoff , S. Werner , G. Ölçücü , U. Schwaneberg , K. E. Jaeger , M. D. Davari , U. Krauss , Machine Learning-Assisted Engineering of Light, Oxygen , Voltage Photoreceptor Adduct Lifetime. JACS Au 3 , 3311 – 3323 ( 2023 ). OpenUrl PubMed 50. ↵ D. Kalderon , B. L. Roberts , W. D. Richardson , A. E. Smith , A short amino acid sequence able to specify nuclear location . Cell 39 , 499 – 509 ( 1984 ). OpenUrl CrossRef PubMed Web of Science 51. ↵ D. D. Jenness , A. C. Burkholder , L. H. Hartwell , Binding of α-factor pheromone to yeast a cells: Chemical and genetic evidence for an α-factor receptor . Cell 35 , 521 – 529 ( 1983 ). OpenUrl CrossRef PubMed Web of Science 52. ↵ A. Bender , G. F. Sprague , Yeast peptide pheromones, a-factor and α-factor, activate a common response mechanism in their target cells . Cell 47 , 929 – 937 ( 1986 ). OpenUrl CrossRef PubMed Web of Science 53. ↵ D. D. Jenness , A. C. Burkholder , L. H. Hartwell , Binding of α-factor pheromone to yeast a cells: Chemical and genetic evidence for an α-factor receptor . Cell 35 , 521 – 529 ( 1983 ). OpenUrl CrossRef PubMed Web of Science 54. ↵ X. Xu , L. Nivon , J. F. Federizon , M. Maglaqui , H. Janjua , L. Mao , R. Xiao , G. Kornhaber , D. Baker , G. T. Montelione , T. Szyperski , Solution NMR Structure of De novo designed protein, Northeast Structural Genomics Consortium (NESG) Target OR462 . Worldwide Protein Data Bank , doi: 10.2210/PDB2MR6/PDB ( 2014 ). OpenUrl CrossRef 55. ↵ P. S. Huang , Y. E. A. Ban , F. Richter , I. Andre , R. Vernon , W. R. Schief , D. Baker , RosettaRemodel: A Generalized Framework for Flexible Backbone Protein Design . PLoS One 6 , e24109 ( 2011 ). OpenUrl CrossRef PubMed 56. ↵ J. L. Watson , D. Juergens , N. R. Bennett , B. L. Trippe , J. Yim , H. E. Eisenach , W. Ahern , A. J. Borst , R. J. Ragotte , L. F. Milles , B. I. M. Wicky , N. Hanikel , S. J. Pellock , A. Courbet , W. Sheffler , J. Wang , P. Venkatesh , I. Sappington , S. V. Torres , A. Lauko , V. De Bortoli , E. Mathieu , S. Ovchinnikov , R. Barzilay , T. S. Jaakkola , F. DiMaio , M. Baek , D. Baker , De novo design of protein structure and function with RFdiffusion . Nature 2023 620 :7976 620, 1089 – 1100 ( 2023 ). OpenUrl CrossRef PubMed 57. ↵ J. Dauparas , I. Anishchenko , N. Bennett , H. Bai , R. J. Ragotte , L. F. Milles , B. I. M. Wicky , A. Courbet , R. J. de Haas , N. Bethel , P. J. Y. Leung , T. F. Huddy , S. Pellock , D. Tischer , F. Chan , B. Koepnick , H. Nguyen , A. Kang , B. Sankaran , A. K. Bera , N. P. King , D. Baker , Robust deep learning–based protein sequence design using ProteinMPNN . Science ( 1979 ) 378 , 49 – 56 ( 2022 ). OpenUrl CrossRef PubMed 58. ↵ M. Mirdita , K. Schütze , Y. Moriwaki , L. Heo , S. Ovchinnikov , M. Steinegger , ColabFold: making protein folding accessible to all . Nature Methods 2022 19 : 6 19, 679 – 682 ( 2022 ). OpenUrl PubMed 59. ↵ J. Jumper , R. Evans , A. Pritzel , T. Green , M. Figurnov , O. Ronneberger , K. Tunyasuvunakool , R. Bates , A. Žídek , A. Potapenko , A. Bridgland , C. Meyer , S. A. A . Kohl , A. J. Ballard , A. Cowie , B. Romera-Paredes , S. Nikolov , R. Jain , J. Adler , T. Back , S. Petersen , D. Reiman , E. Clancy , M. Zielinski , M. Steinegger , M. Pacholska , T. Berghammer , S. Bodenstein , D. Silver , O. Vinyals , A. W. Senior , K. Kavukcuoglu , P. Kohli , D. Hassabis , Highly accurate protein structure prediction with AlphaFold . Nature 2021 596 :7873 596, 583 – 589 ( 2021 ). OpenUrl CrossRef PubMed 60. ↵ A. Leaver-Fay , R. Jacak , P. B. Stranges , B. Kuhlman , A Generic Program for Multistate Protein Design . PLoS One 6 , e20937 ( 2011 ). OpenUrl CrossRef PubMed 61. ↵ P. Szilárd , M. J. Abraham , C. Kutzner , B. Hess , E. Lindahl , Tackling Exascale Software Challenges in Molecular Dynamics Simulations with GROMACS . Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics ) 8759, 3 – 27 ( 2015 ). 62. ↵ M. J. Abraham , T. Murtola , R. Schulz , S. Páll , J. C. Smith , B. Hess , E. Lindah , GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers . SoftwareX 1–2, 19 – 25 ( 2015 ). 63. ↵ J. B. Klauda , R. M. Venable , J. A. Freites , J. W. O’Connor , D. J. Tobias , C. Mondragon-Ramirez , I. Vorobyov , A. D. MacKerell , R. W. Pastor , Update of the CHARMM All-Atom Additive Force Field for Lipids: Validation on Six Lipid Types . Journal of Physical Chemistry B 114 , 7830 – 7843 ( 2010 ). OpenUrl CrossRef 64. ↵ F. Pedregosa FABIANPEDREGOSA , V. Michel , O. Grisel OLIVIERGRISEL , M. Blondel , P. Prettenhofer , R. Weiss , J. Vanderplas , D. Cournapeau , F. Pedregosa , G. Varoquaux , A. Gramfort , B. Thirion , O. Grisel , V. Dubourg , A. Passos , M. Brucher , M. Perrot andÉdouardand, andÉdouard Duchesnay, Fré. Duchesnay EDOUARDDUCHESNAY, Scikit-learn: Machine Learning in Python . Journal of Machine Learning Research 12 , 2825 – 2830 ( 2011 ). OpenUrl 65. ↵ R. D. Gietz , R. H. Schiestl , High-efficiency yeast transformation using the LiAc/SS carrier DNA/PEG method . Nature Protocols 2007 2 : 1 2, 31 – 34 ( 2007 ). OpenUrl PubMed 66. ↵ G. Diss , C. R. Landry , Combining the Dihydrofolate Reductase Protein-Fragment Complementation Assay with Gene Deletions to Establish Genotype-to-Phenotype Maps of Protein Complexes and Interaction Networks . Cold Spring Harb Protoc 2016, pdb.prot090035 ( 2016 ). 67. ↵ J. B. Hicks , I. Herskowitz , Evidence for a new diffusible element of mating pheromones in yeast . Nature 1976 260 :5548 260, 246 – 248 ( 1976 ). OpenUrl CrossRef PubMed 68. ↵ G. F. Sprague , I. Herskowitz , Control of yeast cell type by the mating type locus: I. Identification and control of expression of the a-specific gene BAR1 . J Mol Biol 153 , 305 – 321 ( 1981 ). OpenUrl CrossRef PubMed Web of Science 70. ↵ R. L. J. Keller , The Computer Aided Resonance Assignment Tutorial . ( 2004 ). 70. ↵ P. Klukowski , R. Riek , P. Guntert , NMRtist: an online platform for automated biomolecular NMR spectra analysis . Bioinformatics 39 ( 2023 ). 71. ↵ P. Klukowski , R. Riek , P. Güntert , Rapid protein assignments and structures from raw NMR spectra with the deep learning technique ARTINA . Nature Communications 2022 13 : 1 13, 1 – 12 ( 2022 ). OpenUrl CrossRef PubMed 72. ↵ E. Schmidt , P. Güntert , A new algorithm for reliable and general NMR resonance assignment . J Am Chem Soc 134 , 12817 – 12829 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 73. D. Gottstein , D. K. Kirchner , P. Güntert , Simultaneous single-structure and bundle representation of protein NMR structures in torsion angle space . J Biomol NMR 52 , 351 – 364 ( 2012 ). OpenUrl CrossRef PubMed 74. D. Eisenberg , E. Schwarz , M. Komaromy , R. Wall , Analysis of membrane and surface protein sequences with the hydrophobic moment plot . J Mol Biol 179 , 125 – 142 ( 1984 ). OpenUrl CrossRef PubMed Web of Science 75. E. Conti , M. Uy , L. Leighton , G. Blobel , J. Kuriyan , Crystallographic Analysis of the Recognition of a Nuclear Localization Signal by the Nuclear Import Factor Karyopherin α . Cell 94 , 193 – 204 ( 1998 ). OpenUrl CrossRef PubMed Web of Science 77. V. Velazhahan , N. Ma , G. Pándy-Szekeres , A. J. Kooistra , Y. Lee , D. E. Gloriam , N. Vaidehi , C. G. Tate , Structure of the class D GPCR Ste2 dimer coupled to two G proteins . Nature 2020 589 :7840 589, 148 – 153 ( 2020 ). OpenUrl PubMed View the discussion thread. Back to top Previous Next Posted November 05, 2025. Download PDF Supplementary Material 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 De novo design of light-regulated dynamic proteins using deep learning 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 De novo design of light-regulated dynamic proteins using deep learning L. Scutteri , L.A. Abriata , S. Zhang , A. Hacisuleyman , K. Lau , F. Pojer , S.J. Rahi , P. Barth bioRxiv 2025.08.12.669910; doi: https://doi.org/10.1101/2025.08.12.669910 Share This Article: Copy Citation Tools De novo design of light-regulated dynamic proteins using deep learning L. Scutteri , L.A. Abriata , S. Zhang , A. Hacisuleyman , K. Lau , F. Pojer , S.J. Rahi , P. Barth bioRxiv 2025.08.12.669910; doi: https://doi.org/10.1101/2025.08.12.669910 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 Biophysics Subject Areas All Articles Animal Behavior and Cognition (7618) Biochemistry (17633) Bioengineering (13856) Bioinformatics (41841) Biophysics (21399) Cancer Biology (18529) Cell Biology (25422) Clinical Trials (138) Developmental Biology (13352) Ecology (19860) Epidemiology (2067) Evolutionary Biology (24282) Genetics (15582) Genomics (22462) Immunology (17700) Microbiology (40295) Molecular Biology (17140) Neuroscience (88419) Paleontology (666) Pathology (2823) Pharmacology and Toxicology (4813) Physiology (7632) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4284) Systems Biology (9808) Zoology (2267)