CiRCus: A Framework to Enable Classification of Complex High-Throughput Experiments.

Area of Science:

• Proteomics
• Computational Biology
• Machine Learning

Background:

• High-throughput experiments in molecular biology generate vast datasets.
• Current methods for analyzing these results often require extensive manual effort.
• Automated classification tools are needed to keep pace with experimental output.

Purpose of the Study:

• To present CiRCus, a novel framework for classifying results from high-throughput proteomics binding experiments.
• To develop custom machine learning models for automated data analysis.
• To reduce the manual workload in evaluating experimental outcomes.

Main Methods:

• Development of the CiRCus framework, comprising CindeR for data labeling and CurveClassification for model generation.
• Application of random forest classifiers with optional optimization.
• Training and testing on a dataset of 557,166 protein/drug binding curves.

Main Results:

• Achieved a high classification performance with an Area Under the Curve (AUC) of 0.9987.
• Demonstrated that only 6% of the data may require manual investigation after classification.
• Successfully applied the framework to a large-scale proteomics dataset.

Conclusions:

• CiRCus offers an efficient and accurate solution for classifying high-throughput proteomics binding data.
• The framework significantly minimizes the need for manual data analysis.
• CiRCus enhances the scalability and throughput of molecular biology research.