Peptide Design using RNN

LSTM-Based Deep Learning Model for the Discovery of Antimicrobial Peptides Targeting Mycobacterium tuberculosis

Developed LSTM-based classifiers and generators using transfer learning to predict and generate antimicrobial peptides (AMPs) specific to Mycobacterium tuberculosis.

Benchmarked multiple LSTM architectures, achieving 90% accuracy and 0.97 AUC in TB-AMP classification, with transfer learning enhancing performance under limited data.

Designed a generative pipeline that produced 94 novel non-toxic peptide candidates, 10 of which met physicochemical criteria and structural validation for AMP-like properties.

Performed comparative motif and structural divergence analysis to ensure novelty and function of generated peptides, with full reproducibility provided via public code.


TOAST Image 1

Generated protein structure prediction and visualization