2013/11/14 by Sébastien Giguère, Giguère, Sébastien, François Laviolette +11
Biochemistry, Genetics and Molecular Biology · Immunology and Microbiology · #92B05 #Antimicrobial Peptides and Activities #Biochemical and Structural Characterization #Chemical Synthesis and Analysis #FOS: Biological sciences #G.3 #G.4 #I.2.6 #I.5.2 #J.3 #Quantitative Methods (q-bio.QM) #vaccines and immunoinformatics approaches
paper · pdf · doi:10.48550/arxiv.1311.3573
openalex publication_date 2013/11/14 · openalex created_date 2022/10/04 · openalex updated_date 2026/07/28
The discovery of peptides having high biological activity is very challenging\nmainly because there is an enormous diversity of compounds and only a minority\nhave the desired properties. To lower cost and reduce the time to obtain\npromising compounds, machine learning approaches can greatly assist in the\nprocess and even replace expensive laboratory experiments by learning a\npredictor with existing data. Unfortunately, selecting ligands having the\ngreatest predicted bioactivity requires a prohibitive amount of computational\ntime. For this combinatorial problem, heuristics and stochastic optimization\nmethods are not guaranteed to find adequate compounds.\n We propose an efficient algorithm based on De Bruijn graphs, guaranteed to\nfind the peptides of maximal predicted bioactivity. We demonstrate how this\nalgorithm can be part of an iterative combinatorial chemistry procedure to\nspeed up the discovery and the validation of peptide leads. Moreover, the\nproposed approach does not require the use of known ligands for the target\nprotein since it can leverage recent multi-target machine learning predictors\nwhere ligands for similar targets can serve as initial training data. Finally,\nwe validated the proposed approach in vitro with the discovery of new cationic\nanti-microbial peptides.\n Source code is freely available at\nhttp://graal.ift.ulaval.ca/peptide-design/.\n