From Minimal Data To Maximal Insight: A Machine Learning Guided Platform For Peptide Discovery

Pouriya Bayat, Spencer Perkins, Sebastian Clancy, Sahil Swapnesh Patel, Richard Fei Yin, Krištof Bozovičar, Idorenyin IWE, Mohammad Simchi, Ilan Yaniv Zeisler, Serena Singh, Vivian White, Matthew Xie, Sean Palter, Keith Pardee

Track: Biology: datasets and/or experimental results

Nature Biotechnology: Yes

Keywords: Peptide Discovery, Data-Efficient Algorithms, Computational Biology, Machine Learning, GFP

TL;DR: MDMI uses limited data to design novel, highly divergent peptides, surpassing traditional dataset requirements for peptide discovery.

Abstract: Peptide biologics represent a promising therapeutic frontier, but their discovery and optimization are often hindered by the requirement for extensive training datasets in machine learning approaches. Here we present Minimal Data Maximal Insight (MDMI), a novel computational method that enables peptide discovery using limited data (~100 sequences). Using a split Green Fluorescent Protein (GFP) system as our model, we develop a sequence-agnostic model with statistical potential scoring and physics-based evaluation to create an ensemble predictive model. This is coupled with a genetic algorithm for sequence optimization. With only one round of screening, we developed a model that yielded novel functional sequences 63% of which exhibited fluorescence. Notably, by analyzing high-activity sequences to identify favorable amino acids at each position, we were able to design peptide variants with more than 50% sequence difference from the wild type -far exceeding the mutation rates present in our training data- while maintaining functionality. By reducing dependency on large datasets, MDMI democratizes access to advanced computational tools for peptide engineering and offers a blueprint for accelerating therapeutic peptide discovery across various applications, from antimicrobials to targeted drug delivery.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Pouriya_Bayat1

Format: Maybe: the presenting author will attend in person, contingent on other factors that still need to be determined (e.g., visa, funding).

Funding: Yes, the presenting author of this submission falls under ICLR’s funding aims, and funding would significantly impact their ability to attend the workshop in person.

Submission Number: 97