Benchmarking protein structure predictors to assist machine learning-guided peptide discovery

Abstract

Machine learning models provide an informed and efficient strategy to create novel peptide and protein sequences with the desired profiles. Nevertheless, they are primarily trained on sequences where the tridimensional...

Structure-aware machine learning strategies for antimicrobial peptide discovery

“…Aware of the possible over-representation of α-helical peptides in our models, we evaluated the structural diversity of the peptide datasets. We recently developed a fast and reliable approach to estimate the structural landscape of any sizable peptide dataset, using protein structure predictors, PEP2D and AlphaFold2 (AF2) 29 . Many of the sequences in our datasets contained 50 or more residues, guiding our preferences for AF2.…”

“…We recently recommended a fast and robust estimate of the structural landscape(s) of medium-large datasets for fold discovery and prior ML modeling. Our predictions identified loose α-helices as the main structure class (65.1%), followed by random coils (17.8%), and β-stranded and mixed structures accounted for the rest of the large AMP dataset 29 . Consequently, current AMP models (predictors and generators) might favor these dominant structure classes.…”

“…Aware of the possible over-representation of α-helices among AMPs 29 , we estimated the structural landscapes of our model datasets, revealing a majority of peptides assuming α-helical and loose structures (subset I—53.9%), two minorly represented coiled structures (III—33.1%) and mixed structures (IV—13.0%), and the complete absence of β-stranded peptides (II). An objective evaluation of the structural awareness demonstrated that our preliminary models were biased toward predicting the likelihood of sequences forming α-helical structures by assigning them high-class probabilities.…”

“…We displayed the collective secondary structures of all peptide sequences as single points on a ternary representation—the structural landscape, as previously described 29 . The three axes represent the three secondary structure states—% helix (H), % sheet (E), and % coil (C).…”