Research

For billions of years, nature has been conducting the ultimate experiment. Today—with global sample collection, single-cell metagenomics, and massive sequence datasets—we can finally read the notes.

Our lab develops computational methods, deep learning models, and structural biology tools to explore three core frontiers:

01 • EVOLUTION & SEQUENCE

A unified statistical model of protein evolution — integrating phylogenetic, genomic, structural, and functional constraints.

How do evolutionary constraints shape generative sequence models? Can we disentangle phylogenetic signal from true coevolutionary couplings?

Coevolution, conservation, and phylogenetics: a unified model of protein evolution. — Unified framework: linking conservation, coevolution, and phylogenetic trees.

Learning the MRF (GREMLIN / Potts) through factored and standard attention. — Markov Random Fields & Attention: learning sequence coevolution with Potts models.

Recent Theme Publications

Expanding the scope of protein language modeling to protein-protein interactions with MSA Pairformer. 2026 Akiyama Y, et al. GitHub

Hit or Miss: Understanding Emergence and Absence of Homo-oligomeric Contacts in Protein Language Models. 2025 Zhang Z, et al.

Disentanglement of Evolutionary Constraints in Statistical Models of Proteins. 2024 Wang H, et al. GitHub

Differentiable Search of Evolutionary Trees from Leaves. 2023 Hettiarachchi R, et al. GitHub

Genomic language model predicts protein co-regulation and function. 2023 Hwang Y, et al. GitHub

End-to-end learning of multiple sequence alignments with differentiable Smith–Waterman. 2022 Petti S, et al. GitHub

02 • DIFFUSION & LANDSCAPES

Modeling the protein conformational and folding landscape for structure prediction and design.

Can we generate stable de novo proteins by jointly optimizing sequence & structure landscapes? How can generative diffusion models serve as effective statistical potentials?

The protein conformational landscape: P(sequence | structure) and P(structure | sequence). — Conformational landscape: mapping joint distributions P(sequence | structure) & P(structure | sequence). Image credit: Basile Wicky

Recent Theme Publications

Protein Diffusion Models as Statistical Potentials. 2025 Roney JP, et al. GitHub

Stable de novo protein design via joint conformational landscape and sequence optimization. 2025 Cho Y, et al. GitHub

State-of-the-Art estimation of protein model accuracy using AlphaFold. 2022 Roney JP, et al. GitHub

Structure-based protein design with deep learning. 2021 Ovchinnikov S & Huang PS.

Protein sequence design by conformational landscape optimization. 2021 Norn C, et al. GitHub

03 • METAGENOMICS & DARK MATTER

Mining metagenomic "dark matter" for new protein families, functions, and interactions — and probing the early evolution of multicellularity.

What unknown protein families and functional universe lie in uncultivated metagenomes? How did early molecular complexes evolve to enable multicellular life?

Mining metagenomic dark matter: multiple sequence alignments, structures, and contact maps. — Dark Matter Mining: MSAs, predicted 3D structures, and inter-residue contact maps.

Interaction networks, symbiotic communities, and the evolution of multicellularity. — Symbiotic networks: coevolutionary maps revealing the origins of multicellularity.

Recent Theme Publications

CIRPIN: Learning Circular Permutation-Invariant Representations to Uncover Putative Protein Homologs. 2025 Kolodziej A, et al. GitHub

Unraveling the functional dark matter through global metagenomics. 2023 Pavlopoulos GA, et al. Database

SWAMPNN: End-to-end protein structures alignment. 2022 Trinquier J, et al. GitHub

Protein structure determination using metagenome sequence data. 2017 Ovchinnikov S, et al. GitHub