Building the blocks of life

2026-03-31 · Source: MIT News - Machine learning · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, medium

Summary

Computational biologist Sergei Kotelnikov, a School of Science Dean’s Postdoctoral Fellow at MIT, is developing advanced protein modeling methods using machine learning. Working with the Keating Lab, his research focuses on analyzing and predicting biomolecule structures, particularly proteins, which are fundamental to all organisms. Kotelnikov's work aims to address challenges in understanding how proteins fold and interact, which is crucial for developing medical interventions for diseases caused by misfolded proteins. His prior work at Stony Brook University included creating methods to predict protein complex structures mediated by PROTACs, a new class of molecules targeting previously undruggable proteins in cancer. He achieved this by conceptually dividing flexible linkers and using Fast Fourier Transform algorithms. At MIT, he plans to integrate physics and geometric deep learning into neural network architectures to improve model generalizability and reduce data requirements, with applications in cancer immunology and crop protection.

Key takeaway

For computational biologists and AI scientists working on drug discovery or protein engineering, Kotelnikov's approach of integrating physical and geometric knowledge into neural network architectures offers a path to more generalizable and data-efficient models. You should explore incorporating domain-specific structural constraints and mathematical transformations, like Fast Fourier Transform, into your machine learning pipelines to tackle complex biomolecular interactions, especially for challenging targets like PROTACs.

Key insights

Integrating physics and geometric deep learning enhances protein modeling, improving generalizability and reducing data needs.

Principles

• Protein structure dictates function.
• Emergence of complexity drives biological systems.

Method

Kotelnikov's method for PROTAC modeling involves conceptually cutting flexible linkers into halves, modeling each separately, and then using Fast Fourier Transform for efficient configuration calculation.

In practice

• Model PROTACs by splitting flexible linkers.
• Apply geometric deep learning to biomolecular data.

Topics

• Protein Modeling
• Machine Learning
• Computational Biology
• PROTACs
• Geometric Deep Learning

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• A world model for proteins is here — Biohub's open-source protein "world model" (ESMFold2, ESM Atlas) accelerates drug discovery by mapping, predicting, and designing proteins.
• FreeBridge: Variational Schrödinger Bridges for Cellular Transition Dynamics — Geometric grounding is crucial for biologically interpretable single-cell perturbation dynamics modeling.
• PUFFIN: Protein Unit Discovery with Functional Supervision — PUFFIN discovers protein units by jointly learning structural partitioning and functional supervision via graph neural networks.
• Identification and engineering of highly functional potyviral proteases in cells using co-evolutionary models — Co-evolutionary models can accurately predict and engineer superior potyviral proteases for precise cellular control.
• Structure-Guided Adaptive Propagation for Protein-Protein Interaction Site Prediction — SGAP-PPIS improves protein-protein interaction site prediction by adaptively propagating information based on residue-level geometric microenvironments.
• Integration of alternative fragmentation techniques into standard LC-MS workflows using a single deep learning model enhances proteome coverage — A unified deep learning model and integrated MS platform enhance proteome coverage by leveraging diverse fragmentation techniques.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Machine learning.