Dynamic kernel matching for non-conforming data: A case study of T cell receptor datasets

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Learning

Background:

• Traditional statistical classifiers require structured, tabular data (rows and columns).
• Many biological datasets, such as T-cell receptor (TCR) sequences, lack this conventional structure.
• Analyzing non-conforming data is crucial for uncovering biological patterns and disease signatures.

Purpose of the Study:

• To develop and validate a novel method, dynamic kernel matching (DKM), for adapting statistical classifiers to handle non-conforming data.
• To apply DKM to biological datasets for disease diagnosis.
• To evaluate the performance of DKM-augmented classifiers.

Main Methods:

• Modification of established statistical classifiers using dynamic kernel matching (DKM).
• Application of DKM to two distinct non-conforming datasets: TCR sequences and TCR repertoires.
• Performance evaluation using standard and indeterminant diagnosis metrics on holdout data.

Main Results:

• DKM successfully enabled statistical classifiers to analyze non-conforming TCR sequence and repertoire data.
• The method demonstrated effective performance in identifying disease-related patterns.
• Identified patterns used by classifiers align with existing experimental findings.

Conclusions:

• Dynamic kernel matching (DKM) is a viable approach for analyzing non-conforming data in statistical classification.
• DKM enhances the utility of statistical models for biological sequence data, aiding in disease diagnosis.
• The identified patterns provide biological insights and validate the DKM approach.