DNSS2

DNSS2 predicts protein secondary structure (alpha-helices, beta-strands, and coils) using deep learning to improve accuracy for studies of protein folding and structure-function relationships.

Key Features:

• Advanced Deep Learning Architectures: Integrates six novel one-dimensional deep convolutional, recurrent, residual, memory, fractal, and inception networks specifically tailored for protein secondary structure prediction.
• Enhanced Profile Features: Uses sensitive profile features derived from Hidden Markov Models (HMM) and multiple sequence alignments (MSA) to capture sequence patterns relevant to secondary structure.
• Benchmarking Performance: Benchmarked against two independent test datasets and eight state-of-the-art tools, and in the 2018 CASP13 experiment achieved Q3 = 83.74% and SOV = 72.46%, ranking as the top performer across 82 protein targets.

Scientific Applications:

• Protein folding and structure–function relationships: Provides secondary structure predictions that inform analyses of folding mechanisms and structure–function correlations.
• Protein inter-residue contact prediction: Supplies secondary structure constraints that aid prediction of inter-residue contacts.
• Ab initio tertiary structure prediction: Supplies high-accuracy secondary structure inputs for ab initio modeling of three-dimensional protein conformation from sequence.

Methodology:

Integrates multiple one-dimensional deep neural network architectures (convolutional, recurrent, residual, memory, fractal, inception) and employs profile features derived from Hidden Markov Models (HMM) and multiple sequence alignments (MSA) for secondary-structure prediction.