About

Sergey Ovchinnikov is a Helen and Irwin Sizer Career Development Professor at MIT Department of Biology. He studies protein structure and evolution at environmental, organismal, genomic, structural, and molecular scales. His research employs phylogenetic inference, protein structure prediction and determination, protein design, deep learning, energy-based models, and differentiable programming to address evolutionary questions. His goal is to develop a unified model of protein evolution.

Research topics

• Computer Science
• Artificial Intelligence
• Biology
• Machine Learning
• Biochemistry
• Chemistry
• Computational biology
• Engineering
• Biological system
• Materials science
• Data Mining
• Evolutionary biology
• Genetics
• Algorithm
• Human–computer interaction
• Cell biology
• Programming language
• Systems engineering
• Software engineering
• Nanotechnology
• World Wide Web
• Astronomy
• Data science
• Physics

Selected publications

• Designing Novel Solenoid Proteins with In Silico Evolution

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-04-24 · 4 citations

  preprintOpen access

  Abstract Solenoid proteins are elongated tandem repeat proteins with diverse biological functions, making them attractive targets for protein design. Advances in machine learning have transformed our understanding of sequence-structure relationships, enabling new approaches for de novo protein design. Here, we present an in silico evolution platform that couples a solenoid discriminator network with AlphaFold2 as an oracle within a genetic algorithm. Starting from random sequences, we design α-, β-, and αβ-solenoid backbones, generating structures that span natural and novel solenoid space. We experimentally characterise 41 solenoid designs, with α-solenoids consistently folding as intended, including one structurally validated design that closely matches the design model. All β-solenoids initially failed, reflecting the difficulty of designing β-strand majority proteins. By introducing terminal capping elements and refining designs based on earlier experimental screens, we generate two β-solenoids that have biophysical properties consistent with their designs. Our approach achieves fold-specific hallucination-based design without depending on explicit structural templates.

  PublisherOA PDFDOI

• One-shot design of functional protein binders with BindCraft

  Universität Zürich, ZORA · 2025-08-27

  articleOpen access

  Protein-protein interactions are at the core of all key biological processes. However, the complexity of the structural features that determine protein-protein interactions makes their design challenging. Here we present BindCraft, an open-source and automated pipeline for de novo protein binder design with experimental success rates of 10-100%. BindCraft leverages the weights of AlphaFold2 (ref. ) to generate binders with nanomolar affinity without the need for high-throughput screening or experimental optimization, even in the absence of known binding sites. We successfully designed binders against a diverse set of challenging targets, including cell-surface receptors, common allergens, de novo designed proteins and multi-domain nucleases, such as CRISPR-Cas9. We showcase the functional and therapeutic potential of designed binders by reducing IgE binding to birch allergen in patient-derived samples, modulating Cas9 gene editing activity and reducing the cytotoxicity of a foodborne bacterial enterotoxin. Last, we use cell-surface-receptor-specific binders to redirect adeno-associated virus capsids for targeted gene delivery. This work represents a significant advancement towards a 'one design-one binder' approach in computational design, with immense potential in therapeutics, diagnostics and biotechnology.

  PublisherDOI

• Biomedical data and AI

  Science China Life Sciences · 2025-03-14 · 2 citations

  article
  PublisherDOI

• Scaling down protein language modeling with MSA Pairformer

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-08-03 · 12 citations

  preprintOpen accessSenior authorCorresponding

  Abstract Recent efforts in protein language modeling have focused on scaling single-sequence models and their training data, requiring vast compute resources that limit accessibility. Although models that use multiple sequence alignments (MSA), such as MSA Transformer, offer parameter-efficient alternatives by extracting evolutionary information directly from homologous sequences rather than storing it in parameters, they generally underperform compared to single-sequence-based language due to memory inefficiencies that limit the number of sequences and averaging evolutionary signals across the MSA. We address these challenges with MSA Pairformer, a 111M parameter memory-efficient MSA-based protein language model that extracts evolutionary signals most relevant to a query sequence through bi-directional updates of sequence and pairwise representations. MSA Pairformer achieves state-of-the-art performance in unsupervised contact prediction, outperforming ESM2-15B by 6% points while using two orders of magnitude fewer parameters. In predicting contacts at protein-protein interfaces, MSA Pair-former substantially outperforms all methods with a 24% point increase over MSA Transformer. Unlike single-sequence models that deteriorate in variant effect prediction as they scale, MSA Pairformer maintains strong performance in both tasks. Ablation studies reveal triangle operations remove indirect correlations, and unlike MSA Transformer, MSA Pairformer does not hallucinate contacts after removing covariance, enabling reliable screening of interacting sequence pairs. Overall, our work presents an alternative to the current scaling paradigm in protein language modeling, enabling efficient adaptation to rapidly expanding sequence databases and opening new directions for biological discovery.

  PublisherOA PDFDOI

• Protein Hunter: exploiting structure hallucination within diffusion for protein design

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-10-10 · 2 citations

  preprintOpen accessSenior authorCorresponding

  1 Abstract Interactions between proteins and other biomolecules underlie nearly all biological processes, yet designing such interactions de novo remains challenging. Capturing their specific interactions and co-optimizing sequence and structure are difficult and often require extensive computation. We present Protein Hunter, a fast, fine-tuning-free framework for de novo protein design. Starting from an all-X sequence, we find diffusion-based structure prediction models hallucinate reasonable looking structures that can be further improved through iterative sequence re-design and structure re-prediction. This lightweight strategy achieves high AlphaFold3 in silico success rates across both unconditional and conditional generation tasks, including binders to proteins, cyclic peptides, small molecules, DNA, and RNA. Protein Hunter also supports multi-motif scaffolding and partial redesign, providing a general and efficient platform for de novo protein design across diverse molecular targets.

  PublisherOA PDFDOI

• Learning millisecond protein dynamics from what is missing in NMR spectra

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-03-19 · 10 citations

  preprintOpen access

  Abstract Many proteins’ biological functions rely on interconversions between multiple conformations occurring at micro- to millisecond (µs-ms) timescales. A lack of standardized, large-scale experimental data has hindered obtaining a more predictive understanding of these motions. After curating >100 Nuclear Magnetic Resonance (NMR) relaxation datasets, we realized an observable for µs-ms dynamics might be hiding in plain sight. Millisecond dynamics can cause NMR signals to broaden beyond detection, leaving some residues not assigned in the chemical shift datasets of ∼10,000 proteins deposited in the Biological Magnetic Resonance Data Bank (BMRB) 1 . We made the bold assumption that residues missing assignments are exchange-broadened due to µs-ms motions and trained various deep learning models to predict missing assignments. Strikingly, these models also predict exchange measured via NMR relaxation experiments, indicative of µs-ms dynamics. The best of these models, which we named Dyna-1, leverages an intermediate layer of the multimodal language model ESM-3 2 . Notably, dynamics directly linked to biological function — including enzyme catalysis and ligand binding — are particularly well predicted by Dyna-1, which parallels our findings that residues experiencing µs-ms exchange are more conserved. We anticipate the datasets and models presented here will be transformative in unlocking the common language of dynamics and function.

  PublisherOA PDFDOI

• De novo Design of a Peptide Modulator to Reverse Sodium Channel Dysfunction Linked to Cardiac Arrhythmias and Epilepsy

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-04-28

  preprintOpen access

  Summary Ion channels orchestrate electrical signaling in excitable cells. In nature, ion channel function is customized by modulatory proteins that have evolved to fulfill distinct physiological needs. Yet, engineering synthetic modulators that precisely tune ion channel function is challenging. One example involves the voltage-gated sodium (Na V ) channel that initiates the action potential, and whose dysfunction amplifies late/persistent sodium current ( I NaL ), a commonality that underlies various human diseases including cardiac arrhythmias and epilepsy. Here, using a computational protein design platform, we engineered a de novo peptide modulator, ELIXIR, that binds Na V channels with submicromolar affinity. Functional analysis revealed an unexpected selectivity in inhibiting ‘pathogenic’ I NaL and confirmed its effectiveness in reversing Na V dysfunction linked to both cardiac arrhythmias and epilepsy in cellular and murine models. These findings exemplify the efficacy of de novo protein design for engineering synthetic ion channel modulators and sets the stage for rational design of future therapeutic approaches.

  PublisherOA PDFDOI

• actifpTM: a refined confidence metric of AlphaFold2 predictions involving flexible regions

  Bioinformatics · 2025-03-01 · 42 citations

  articleOpen access

  SUMMARY: One of the main advantages of deep learning models of protein structure, such as Alphafold2, is their ability to accurately estimate the confidence of a generated structural model, which allows us to focus on highly confident predictions. The ipTM score provides a confidence estimate of interchain contacts in protein-protein interactions. However, interactions, in particular motif-mediated interactions, often also contain regions that remain flexible upon binding. These noninteracting flanking regions are assigned low confidence values and will affect ipTM, as it considers all interchain residue-residue pairs, and two models of the same motif-domain interaction, but differing in the length of their flanking regions, would be assigned very different values. Here, we propose actual interface pTM (actifpTM), a modified ipTM measure, that focuses on the residues participating in the interaction, resulting in a more robust measure of interaction confidence. Besides, actifpTM is calculated both for the full complex as well as for each pair of chains, making it well-suited for evaluating multi-chain complexes with a particularly critical binding interface, such as antibody-antigen interactions. AVAILABILITY AND IMPLEMENTATION: The method is available as part of the ColabFold (https://github.com/sokrypton/ColabFold) repository, installable both locally or usable with Colab notebook.

  PublisherOA PDFDOI

• Accurate de novo design of high-affinity protein-binding macrocycles using deep learning

  JuSER Publikationsportal · 2025-01-01

  articleOpen access

  Developing macrocyclic binders to therapeutic proteins typically relies on large-scale screening methods that are resource intensive and provide little control over binding mode. Despite progress in protein design, there are currently no robust approaches for de novo design of protein-binding macrocycles. Here we introduce RFpeptides, a denoising diffusion-based pipeline for designing macrocyclic binders against protein targets of interest. We tested 20 or fewer designed macrocycles against each of four diverse proteins and obtained binders with medium to high affinity against all targets. For one of the targets, Rhombotarget A (RbtA), we designed a high-affinity binder (Kd < 10 nM) despite starting from the predicted target structure. X-ray structures for macrocycle-bound myeloid cell leukemia 1, γ-aminobutyric acid type A receptor-associated protein and RbtA complexes match closely with the computational models, with a Cα root-mean-square deviation < 1.5 Å to the design models. RFpeptides provides a framework for rapid and custom design of macrocyclic peptides for diagnostic and therapeutic applications.

  PublisherDOI

• Boltzdesign1: Inverting All-Atom Structure Prediction Model for Generalized Biomolecular Binder Design

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-04-06 · 21 citations

  preprintOpen accessSenior authorCorresponding

  Abstract Deep learning in structure prediction has revolutionized protein research, enabling large-scale screening, novel hypothesis generation, and accelerated experimental design across biological domains. Recent advances, including RoseTTAFold-AA and AlphaFold3, have extended structure prediction models to work with small molecules, nucleic acids, ions, and covalent modifications. We present BoltzDesign1, which inverts the Boltz-1 model, an open source reproduction of AlphaFold3, to enable the design of protein binders for diverse molecular targets without requiring model finetuning. By utilizing only the Pairformer and Confidence modules, our method significantly reduces computational costs while achieving outstanding in silico success rates and diversity in binder generation. Optimizing directly on the distogram allows us to shape the probability distribution of atomic distances, rather than adjusting a single structure, steering the design toward sequences that yield robust structures with well-defined energy minima. By leveraging a fully atomic model trained on a wide variety of macromolecules, we can generate diverse heterocomplexes with flexible ligand conformations—a capability not currently matched by existing methods. This approach enables the design of novel protein interactions with potential applications in biosensors, enzyme engineering, therapeutic development, and biotechnological innovations.

  PublisherOA PDFDOI

Recent grants

• Exploring the unknown protein universe using evolutionary information

  NIH · $2.0M · 2018–2024

Frequent coauthors

• David Baker

  University of Washington

  108 shared

• Milot Mirdita

  Seoul National University

  33 shared

• Martin Steinegger

  Seoul National University

  32 shared

• Yoshitaka Moriwaki

  The University of Tokyo

  30 shared

• Lim Heo

  Michigan State University

  29 shared

• Justas Dauparas

  University of Washington

  28 shared

• Konstantin Schütze

  Seoul National University

  28 shared

• Ivan Anishchenko

  27 shared

Labs

• Sergey Ovchinnikov LabPI

Education

• PhD, Molecular and Cellular Biology / Biochemistry

  University of Washington

  2017

• BS, Micro/Molecular Biology

  Portland State University

  2010