Unsupervised encoding selection through ensemble pruning for biomedical classification.

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Drug Discovery

Background:

• Rising multi-resistant pathogens necessitate novel antimicrobial strategies beyond traditional antibiotics.
• Antimicrobial peptides (AMPs) offer a promising alternative, but their identification and validation are costly and time-consuming.
• Artificial intelligence (AI) is increasingly used to automate the detection and functional prediction of AMPs.

Purpose of the Study:

• To develop an automated workflow for unsupervised selection of peptide encodings and generation of ensemble classifiers.
• To improve the prediction accuracy of peptide functions, including antimicrobial and other properties.
• To address the lack of a comprehensive unsupervised approach for encoding selection across various machine learning models.

Main Methods:

• Developed a novel workflow integrating automated encoding selection and ensemble classifier generation.
• Employed sophisticated pruning methods, including Pareto frontier pruning, for creating effective encoding ensembles.
• Evaluated various combinations of peptide encodings and base machine learning models, such as Decision Tree classifiers.

Main Results:

• The workflow successfully automates encoding selection and ensemble creation for peptide classification tasks.
• Pareto frontier pruning proved effective for generating robust encoding ensembles across different datasets.
• Ensembles combining specific encodings with Decision Tree classifiers frequently yielded superior predictive performance, though no single ensemble technique was universally optimal.

Conclusions:

• The developed workflow enables unsupervised encoding selection and ensemble classifier evaluation for peptide analysis.
• This tool empowers researchers to efficiently identify and validate promising peptide candidates for therapeutic applications.
• The extensible workflow integrates with existing platforms like PEPTIDE REACToR, enhancing its utility in peptide research.