top of page

Designing antimicrobial peptides against infectious diseases

#Machine Learning #Ligand-Based Drug Design #Peptide Design #Infectious Diseases #Antimicrobial Resistance

Context

Antimicrobial peptides (AMPs) represent a promising class of therapeutic agents to combat antibiotic-resistant infections. However, many candidate peptides suffer from hemolytic toxicity, limited bioavailability, and poor predictive modeling when designed computationally. As a principal investigator and now as Founder of Ingenie Bio, I developed and refined a suite of machine learning pipelines to design safe and structurally diverse peptides targeting infectious pathogens.


Client/Partner Type

Academic Research (with biotech translation relevance)


Challenge

Traditional AMP discovery pipelines struggle with:

  • Poor generalization to novel sequences

  • Bias toward α-helical peptides

  • Inadequate representation of sequence-structure-function relationships

  • Lack of outlier and uncertainty detection in predictive models


The overarching goal was to develop ML strategies that not only predict biological activity and safety, but also define reliable applicability domains and integrate structural diversity — key for real-world infectious disease targeting.


Solution

Over a multi-project program spanning three major publications, we developed a comprehensive, multi-stage framework:


(1) Toxicity-Aware Classification with Outlier Detection

  • Developed ML models to predict hemolytic toxicity, a major barrier to AMP clinical use

  • Trained 14 classifiers (e.g., GBC, SVM, XGBoost) using physicochemical descriptors from curated datasets

  • Integrated 9 outlier detection methods (e.g., LOF, Mahalanobis) to define the applicability domain

  • Used models to screen 3,000+ natural AMPs and generate 500+ low-toxicity peptide designs


Plisson, F.*; Ramírez-Sánchez, O. & Martínez-Hernández, C. Machine learning-guided discovery and design of non-hemolytic peptides. Scientific Reports 2020, 10, 16581. [DOI] [Github]


(2) Structure-Aware ML Modelling

  • Applied secondary structure prediction (PEP2D) to 5,800+ AMPs from GRAMPA

  • Visualized fold space with ternary plots (Helix-Strand-Coil)

  • Identified structural biases in public datasets and AMP design models

  • Combined structural projections with outlier detection to guide more diverse and robust peptide design


Aguilera-Puga, M. d. C. & Plisson, F.*. Structure-aware machine learning strategies for antimicrobial peptide discovery. Scientific Reports 2024, 14, 11995. [DOI] [Github]


(3) Protocolisation and Knowledge Transfer

  • Published a peer-reviewed protocol for researchers to build and evaluate their AMP classifiers

  • Covered data curation, feature engineering, model training, validation, and applicability domain analysis

  • Positioned the approach as a standardised framework for safer and more effective AMP discovery


Aguilera-Puga, M. d. C.; Cancelarich, N. L.; Marani, M. M.; de la Fuente-Nuñez, C.* & Plisson, F.*. Accelerating the discovery and design of antimicrobial peptides with artificial intelligence. Methods in Molecular Biology 2024; 2714 : 329-352. [DOI]


Outcome

  • Discovered 34 natural AMPs with low hemolytic potential from public data

  • Designed 507 de novo peptides with minimised hemolytic risk

  • Defined new structural guidelines and modeling practices to mitigate taxonomic and fold bias

  • Developed a reproducible, extensible modeling pipeline for use in peptide drug discovery

  • Published in leading journals (Scientific Reports) and included in practical Springer Methods


Related articles


Robles-Loaiza, A. A.; Pinos-Tamayo, E. A.; Mendes, B.; Ortega-Pila, J. A.; Proaño-Bolaños, C.; Plisson, F.; Teixeira, C.; Gomes, P.; Almeida, J. R.*. Traditional and Computational Screening of Non-Toxic Peptides and Approaches to Improving Selectivity. Pharmaceuticals 2022, 15, 323. [DOI]


Bajorath, J.; Chávez-Hernández, A. L.; Duran-Frigola, M.; Fernández-de Gortari, E.; Gasteiger, J.; López-López, E.; Maggiora, G. M.; Medina-Franco, J. L.*; Méndez-Lucio, O.; Mestres, J.; Miranda-Quintana, R. A.; Oprea, T. I.; Plisson, F.; Prieto-Martínez, F. D.; Rodríguez-Pérez, R.; Rondón-Villarreal, P.; Saldívar-Gonzalez, F. I.; Sánchez-Cruz, N.; Valli, M. Chemoinformatics and artificial intelligence colloquium: progress and challenges to develop bioactive compounds. Journal of Chemoinformatics 2022, 14(1), 82. [DOI]


Robles-Ramirez, O.; Osuna, J. G.; Plisson, F.; Barrientos-Salcedo; C.*. Antimicrobial Peptides in Livestock: A Review with a One Health Approach. Frontiers in Cellular and Infection Microbiology 2024; vol. 14. [DOI]


Tools/Expertise Used

Python (Scikit-learn, modlAMP, PyOD), Gradient Boosting, Outlier Detection, Feature Engineering, Sequence Analysis, Secondary Structure Prediction (PEP2D), AlphaFold2.


N.B. This work laid the foundation for Ingenie Bio’s service offerings in ML-guided predictions and generation in protein design.

bottom of page