Daniil Litvinov, Lorenzo Pantolini, Peter Škrinjar, Gerardo Tauriello, Caitlyn L. McCafferty, Benjamin D. Engel, Torsten Schwede, Janani Durairaj

Protein complexes are central to cellular function, but experimental determination of their structures remains challenging. Structure prediction methods require prior knowledge of stoichiometry - the number of copies of each protein entity within a complex. Current approaches rely on computationally expensive brute-force methods that run structure prediction on multiple stoichiometry combinations, often with limited accuracy. We introduce Stoic, a method that uses protein language model embeddings to predict protein complex stoichiometry. Our approach learns to identify interface residues that participate in protein-protein interactions, rather than relying on global sequence features. By integrating these interface-aware embeddings into a graph neural network, Stoic achieves fast and accurate stoichiometry prediction for both homomeric and heteromeric targets.

Lorenzo Pantolini, Janani Durairaj

De novo protein design expands the functional protein universe beyond natural evolution, offering vast therapeutic and industrial potential. Monte Carlo sampling in protein design is under-explored due to the typically long simulation times required or prohibitive time requirements of current structure prediction oracles. Here we make use of a 20-letter structure-inspired alphabet derived from protein language model embeddings to score random mutagenesis-based Metropolis sampling of amino acid sequences. This facilitates fast template-guided and unconditional design, generating sequences that satisfy in silico designability criteria without known homologues. Ultimately, this unlocks a new path to fast and de novo protein design.

Peter Škrinjar, Jérôme Eberhardt, Gabriel Studer, Gerardo Tauriello, Torsten Schwede, Janani Durairaj

Deep learning has driven major breakthroughs in protein structure prediction, however the next critical advance is accurately predicting how proteins interact with other molecules, especially small molecule ligands, to enable real-world applications such as drug discovery and design. Recent deep learning all-atom methods have been built to address this challenge, but evaluating their performance on the prediction of protein-ligand complexes has been inconclusive due to the lack of relevant benchmarking datasets. Here we present a comprehensive evaluation of four leading all-atom cofolding deep learning methods using our newly introduced benchmark dataset Runs N' Poses, which comprises 2,600 high-resolution protein-ligand systems released after the training cutoff used by these methods. We demonstrate that current co-folding approaches largely memorise ligand poses from their training data, hindering their use for de novo drug design.

Lorenzo Pantolini, Gabriel Studer, Laura Engist, Ieva Pudziuvelytė, Florian Pommerening, Andrew Mark Waterhouse, Gerardo Tauriello, Martin Steinegger, Torsten Schwede, Janani Durairaj

Detecting remote homology with speed and sensitivity is crucial for tasks like function annotation and structure prediction. We introduce a novel approach using contrastive learning to convert protein language model embeddings into a new 20-letter alphabet, TEA, enabling highly efficient large-scale protein homology searches. Searching with our alphabet performs on par with and complements structure-based methods without requiring any structural information, and with the speed of sequence search. Ultimately, we bring the exciting advances in protein language model representation learning to the plethora of sequence bioinformatics algorithms developed over the past century, offering a powerful new tool for biological discovery.

Janani Durairaj, Yusuf Adeshina, Zhonglin Cao, Xuejin Zhang, Vladas Oleinikovas, Thomas Duignan, Zachary McClure, Xavier Robin, Gabriel Studer, Daniel Kovtun, Emanuele Rossi, Guoqing Zhou, Srimukh Veccham, Clemens Isert, Yuxing Peng, Prabindh Sundareson, Mehmet Akdel, Gabriele Corso, Hannes Stärk, Gerardo Tauriello, Zachary Carpenter, Michael Bronstein, Emine Kucukbenli, Torsten Schwede, Luca Naef

Protein-ligand interactions (PLI) are foundational to small molecule drug design. With computational methods striving towards experimental accuracy, there is a critical demand for a well-curated and diverse PLI dataset. Existing datasets are often limited in size and diversity, and commonly used evaluation sets suffer from training information leakage, hindering the realistic assessment of method generalization capabilities. To address these shortcomings, we present PLINDER, the largest and most annotated dataset to date, comprising 449,383 PLI systems, each with over 500 annotations, similarity metrics at protein, pocket, interaction and ligand levels, and paired unbound (apo) and predicted structures. We propose an approach to generate training and evaluation splits that minimizes task-specific leakage and maximizes test set quality, and compare the resulting performance of DiffDock when retrained with different kinds of splits.