Machine learning algorithms for mode-of-action classification in toxicity assessment.

Area of Science:

• Toxicology
• Computational Biology
• Biotechnology

Background:

• Real Time Cell Analysis (RTCA) enables continuous monitoring of cellular responses to chemical exposures.
• RTCA profiles across various concentrations offer insights into the mode of action (MOA) of substances.
• Accurate MOA determination is crucial for chemical safety assessment and high-throughput screening.

Purpose of the Study:

• To develop and validate machine learning approaches for automated MOA assessment using RTCA data.
• To investigate the efficacy of artificial neural network (ANN) and support vector machine (SVM) algorithms for MOA classification.
• To introduce a novel data processing technique for improved feature extraction from time-concentration response curves (TCRCs).

Main Methods:

• Development of computational tools utilizing ANN and SVM to analyze TCRCs from human cell lines.
• Implementation of wavelet transform for extracting salient features from raw TCRC data.
• Supervised learning applied to HepG2 cell line exposure data for 63 chemicals across 11 concentrations.

Main Results:

• Machine learning models successfully classified MOA based on TCRC data.
• Wavelet transform preprocessing significantly improved feature extraction and model performance.
• SVM achieved classification success rates of 85-95% for MOA classification into two to four clusters.
• Optimized time intervals within dose-response curves enhanced classification accuracy, especially with limited data.

Conclusions:

• Wavelet transform effectively captures critical TCRC features for MOA classification.
• The SVM approach combined with wavelet transform shows significant potential for large-scale MOA classification.
• This methodology is promising for high-throughput chemical screening and toxicological assessments.