HH-suite

HH-suite performs sensitive protein sequence similarity searches and protein fold recognition by pairwise alignment of profile Hidden Markov Models (HMMs) representing multiple sequence alignments.

Key Features:

• Profile HMM Alignment: Uses pairwise alignment of profile Hidden Markov Models (HMMs) to represent multiple sequence alignments of homologous proteins and detect remote homologues.
• Speed Enhancements: Implements a SIMD-vectorized Viterbi algorithm to accelerate profile HMM alignments, with HHsearch accelerated ~4×, HHblits ~2× over previous versions, and HHblits3 reported ~10× faster than PSI-BLAST and ~20× faster than HMMER3.
• Parallelization Capabilities: Supports OpenMP for multi-core systems and MPI for cluster servers to parallelize large-scale searches involving many query profile HMMs.
• Integration with Databases: Integrates with Uniclust90, Uniclust50, and Uniclust30 (clustered protein sequence data at varying pairwise identity levels) with MSAs and functional annotations enriched via MMseqs2 and HHblits.
• Improved Sensitivity and Alignment Quality: Incorporates predicted secondary structure into HMMs to detect more homologous relationships and produce higher-quality alignments with increased balanced scores across family, superfamily, and fold levels compared to PSI-BLAST, HMMER, COMPASS, and PROF_SIM.

Scientific Applications:

• Protein Structure Prediction: Large-scale fold recognition and protein structure prediction in genomics and metagenomics projects.
• Function Prediction and Annotation: Generating MSAs and functional annotations to support protein function inference.
• Homology Detection and Evolutionary Analysis: Sensitive detection of remote homologues to infer evolutionary relationships across family, superfamily, and fold levels.

Methodology:

Pairwise alignment of profile HMMs; SIMD-vectorized implementation of the Viterbi algorithm; incorporation of predicted secondary structure into HMMs; parallelization via OpenMP and MPI; enrichment of Uniclust MSAs and annotations using MMseqs2 and HHblits.