Probe Efficient Feature Representation of Gapped K-mer Frequency Vectors from Sequences Using Deep Neural Networks

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Gapped k-mers frequency vectors (gkm-fvs) are effective for sequence feature extraction.
• Gkm-SVM achieves good predictions but struggles with large datasets due to computational costs.
• High-dimensional gkm-fvs pose challenges for traditional machine learning methods.

Purpose of the Study:

• To develop a flexible and scalable framework (gkm-DNN) for feature representation from gkm-fvs using deep neural networks.
• To address the computational and dimensionality challenges associated with gkm-fvs.
• To enable efficient sequence-based predictions on large biological datasets.

Main Methods:

• Proposed a concise version of gkm-fvs to reduce dimensionality.
• Developed an efficient method for calculating gkm-fvs.
• Implemented a deep neural network (DNN) model utilizing gkm-fvs as input.

Main Results:

• gkm-DNN provides efficient feature representation from high-dimensional gkm-fvs.
• The framework is scalable and handles large datasets effectively.
• Demonstrated performance on transcription factor binding site prediction using ENCODE ChIP-seq data.

Conclusions:

• gkm-DNN offers a scalable and efficient alternative to gkm-SVM for sequence analysis.
• The proposed methods enhance the utility of gkm-fvs in machine learning applications.
• This framework advances sequence-based prediction tasks in computational biology.