Exploring zero-shot structure-based protein fitness prediction

2025-04-23

AI Summary

Computational structural predictions often introduce bias in fitness prediction for variants in intrinsically disordered regions (IDRs), undermining reliability. Method: We propose a zero-shot, fine-tuning-free multimodal framework that jointly leverages pretrained protein language models (e.g., ESM) and AlphaFold2-predicted structures—while critically emphasizing the necessity of experimentally determined structures for accurate modeling. A simple yet effective multimodal integration enables sequence–structure co-modeling. Contributions/Results: (1) We systematically demonstrate that computational structure prediction exacerbates bias in IDRs and establish that high-resolution experimental structures are decisive for robust fitness prediction; (2) we identify structural quality and local disorder-to-order propensity as key determinants of zero-shot performance; (3) on the ProteinGym benchmark, our lightweight multimodal ensemble establishes a strong, plug-and-play zero-shot baseline—enabling reliable interpretation of genetic variants and supporting rational protein engineering without task-specific adaptation.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

The ability to make zero-shot predictions about the fitness consequences of protein sequence changes with pre-trained machine learning models enables many practical applications. Such models can be applied for downstream tasks like genetic variant interpretation and protein engineering without additional labeled data. The advent of capable protein structure prediction tools has led to the availability of orders of magnitude more precomputed predicted structures, giving rise to powerful structure-based fitness prediction models. Through our experiments, we assess several modeling choices for structure-based models and their effects on downstream fitness prediction. Zero-shot fitness prediction models can struggle to assess the fitness landscape within disordered regions of proteins, those that lack a fixed 3D structure. We confirm the importance of matching protein structures to fitness assays and find that predicted structures for disordered regions can be misleading and affect predictive performance. Lastly, we evaluate an additional structure-based model on the ProteinGym substitution benchmark and show that simple multi-modal ensembles are strong baselines.

Problem

• Zero-shot prediction of protein fitness changes using pre-trained models
• Impact of protein structure modeling on fitness prediction accuracy
• Challenges in assessing fitness within disordered protein regions

Innovation

• Zero-shot protein fitness prediction with pre-trained models
• Structure-based models using predicted protein structures
• Multi-modal ensembles for improved predictive performance