TSVM: Transfer Support Vector Machine for Predicting MPRA Validated Regulatory Variants.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) frequently identify common genetic variants in non-coding regions linked to complex diseases.
• Many non-coding variants lack functional validation, hindering the development of predictive models for causal variants.
• Accurate prediction of non-coding causal variants is essential for advancing disease research.

Purpose of the Study:

• To develop a novel transfer learning-based machine learning method for enhanced prediction of regulatory variants.
• To address the challenge of limited sample sizes in predicting non-coding functional variants.
• To improve the accuracy of identifying causal variants in non-coding genomic regions.

Main Methods:

• A transfer support vector machine (TSVM) model was developed for predicting regulatory variants validated by massively parallel reporter assays (MPRA).
• Convolutional neural networks (CNNs) were employed for feature extraction utilizing transfer learning.
• Feature selection was performed using the random forest method, followed by classification with a support vector machine (SVM).

Main Results:

• The transfer learning approach demonstrated effectiveness on the MPRA dataset, as confirmed by scale sensitivity experiments.
• The developed TSVM model achieved a Matthews correlation coefficient (MCC) of 0.326 and an area under the curve (AUC) of 0.720.
• Performance metrics surpassed those of existing state-of-the-art methods.

Conclusions:

• The proposed transfer learning-based machine learning method significantly enhances the prediction accuracy of regulatory variants.
• This approach effectively overcomes the limitations posed by small sample sizes in functional variant prediction.
• The findings offer a promising strategy for identifying non-coding causal variants relevant to complex diseases.