Massachusetts Institute of Technology
Talk Session: SESSION 10: NEW HORIZONS IN PEPTIDE SCIENCE: GREEN METHODS AND DATA SCIENCE
Date: Wednesday, June 15, 2022
Talk Time: 09:35 am - 09:50 am
Talk Title: Targeted Affinity Selection of Peptide Binders Using Machine Learning
Joseph Brown is a chemical biology postdoc in the Pentelute Lab and a member of the CADI board. He grew up in Pilot Mountain, NC and did his undergrad at NC State University. Joe graduated with his Ph.D. from Cornell University and joined the Pentelute Lab in 2019. Joe’s research focuses on expanding beyond the twenty canonical amino acids to make potent protein and peptidomimetic therapeutics.
Affinity selection-mass spectrometry, AS-MS is a widely used technique for the discovery of high-affinity binding molecules to biomolecular targets. The use of large combinatorial libraries has improved the potential of de novo peptide discovery, but remains significantly limited by the capacity of tandem sequencing of affinity selection samples even with high-resolution Orbitrap spectrometers.
With 2 x 108- membered libraries in affinity selection, we observe unbiased sequencing results in high-fidelity sequence assignment of only ~0.2% of peptide features in affinity selections samples. Moreover, non-specific binders are simultaneously sampled, requiring individual analysis and validation of each identified sequence.
In this work, we demonstrate a highly-efficient, targeted sequencing workflow that is coupled with machine learning, ML, to significantly advance the discovery capability of AS-MS using peptide libraries. For proof of concept, we use canonical L-peptide libraries, 2.4 x 109 members total, and multiple target proteins including an anti-hemagglutinin antibody and Mouse double minute 2 homolog, MDM2. Individual features of affinity-selected peptides are autonomously compared to identify and rank highly-enriched, target-specific features for robust tandem sequencing, greatly expanding the number of true putative binders discovered.
With these identified sequences, topological representations of the peptides enable robust encoding for machine learning. Unsupervised learning then distinguishes between the populations of target-specific peptide binders, allowing one to navigate the boundaries and binder families in the peptide sequence space. These efforts will streamline the selection-based identification of target-specific binding molecules and find immediate utility as a powerful tool for drug discovery.