AptaNet as a deep learning approach for aptamer-protein interaction prediction.

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Aptamers are oligonucleotide or peptide molecules with high specificity and affinity for target binding.
• Aptamers show significant potential in biosensing, diagnostics, and therapeutics.
• Predicting aptamer-protein interactions is crucial for their application development.

Purpose of the Study:

• To develop AptaNet, a deep neural network model for predicting aptamer-protein interactions.
• To integrate diverse features from both aptamers and proteins for accurate prediction.
• To provide a valuable tool for identifying novel aptamer-protein pairs.

Main Methods:

• Aptamers were encoded using k-mer and reverse complement k-mer frequencies.
• Proteins were represented using Amino Acid Composition (AAC) and Pseudo Amino Acid Composition (PseAAC) with 24 properties.
• A neighborhood cleaning algorithm addressed data imbalance.
• A deep neural network was employed, with feature selection via random forest.

Main Results:

• AptaNet achieved 99.79% accuracy on the training dataset.
• The model obtained 91.38% accuracy on the independent testing dataset.
• High performance was demonstrated on a constructed aptamer-protein benchmark dataset.

Conclusions:

• AptaNet effectively predicts aptamer-protein interactions.
• The model can facilitate the identification of novel aptamer-protein pairs.
• AptaNet offers insights into aptamer-protein relationships, aiding in biosensing and therapeutic development.