Sparse Markov chain-based semi-supervised multi-instance multi-label method for protein function prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Automated protein function assignment is crucial for analyzing large-scale genomic data.
• Manual annotation is becoming unfeasible due to the rapid increase in sequencing data.
• Existing computational methods face challenges in predicting protein functions accurately.

Purpose of the Study:

• To develop an effective computational method for automated protein function prediction.
• To address the complexity of protein function prediction using a multi-instance multi-label (MIML) framework.
• To propose a novel semi-supervised MIML method leveraging sparse Markov chains.

Main Methods:

• The proposed Sparse-Markov method utilizes a sparse Markov chain-based approach.
• A sparse transductive probability graph is constructed using ensemble Hausdorff distance metrics.
• The method exploits data affinity within the graph to predict protein functions.

Main Results:

• Sparse-Markov demonstrated superior performance compared to four state-of-the-art MIML algorithms.
• Experiments were conducted on seven real-world organism datasets across three biological domains.
• The method effectively assigns similar functional labels to proteins with close affinities.

Conclusions:

• The Sparse-Markov method offers an improved approach for automated protein function prediction.
• The multi-instance multi-label framework combined with sparse Markov chains is effective for this task.
• This computational strategy aids in annotating large genomics datasets.