CycDelta: Multimodal Residue-Graph Learning of Permeability Changes in Cyclic Peptides
Membrane permeability remains a major challenge in cyclic-peptide drug discovery, where optimization often depends on estimating changes relative to measured reference compounds. Existing predictors mainly model absolute permeability, while heterogeneity across literature sources complicates pooled training and comparative evaluation. We present CycDelta, a multimodal residue-graph model trained…
Dr. <unnamed> introduces CycDelta, a novel multimodal residue-graph model designed to predict changes in the permeability of cyclic peptides. Unlike existing predictors that focus on absolute permeability, CycDelta hones in on the differences between measured reference compounds, addressing the complexity and heterogeneity in the literature.
By leveraging directed message passing, pretrained Uni-Mol monomer embeddings, physicochemical descriptors, and assay information, CycDelta offers a comprehensive encoding system for both reference and query peptides. To assess CycDelta's performance, the authors introduce CycDeltaBench, a benchmark encompassing various scenarios, including in-distribution, scaffold-based out-of-distribution, external, and zero-shot unseen-assay tests.
The results are impressive: CycDelta outperforms the strongest baseline by a Pearson correlation of 0.69 compared to 0.46, and it leads in two external datasets, achieving a Spearman correlation of 0.70 versus 0.32 for the benchmark's best performer on an unseen assay. CycDelta's ability to provide residue-level attribution and support gradient-guided screening for permeability-enhancing analogue hypotheses makes it a transferable and interpretable tool for comparative permeability prediction and cyclic-peptide optimization.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
