PolyAMotif

PolyAMotif predicts the 12 main variants of human polyadenylation motifs, including AATAAA, ATTAAA, AAAAAG, AAGAAA, TATAAA, AATACA, AGTAAA, ACTAAA, GATAAA, CATAAA, AATATA, and AATAGA, to identify polyadenylation signals involved in mRNA 3' end formation and stability.

Key Features:

• Machine Learning Integration: Combines generative hidden Markov models (HMMs) with discriminative support vector machines (SVMs) to model sequence uncertainty and optimize classification.
• Spectral Feature Extraction: Uses a spectral algorithm to extract latent-variable features from HMMs for input to SVM classifiers.
• Performance Metrics: Evaluated on 14,740 human poly(A) motif samples, reporting reductions in average error rates (26%), false-negative rates (15%), and false-positive rates (35%), and ~30% fewer erroneous predictions than string kernel methods.
• Visualization Capabilities: Assesses the importance of oligomers and positional signals around motifs, revealing characteristic differences in regions surrounding true and false motifs.

Scientific Applications:

• Genome Annotation: Precise identification of poly(A) signals for annotation of 3' UTRs and transcript ends.
• mRNA Stability and Regulation: Elucidation of regulatory mechanisms that influence mRNA stability via polyadenylation signal detection.
• Transcriptomics and Gene Expression Studies: Support for analyses requiring accurate poly(A) motif mapping in gene expression and transcriptome profiling.
• High-throughput Genomic Analyses: Improvement of motif-calling accuracy in large-scale sequencing datasets.

Methodology:

Analyzes sequential and structural information from DNA sequences surrounding candidate poly(A) motifs using generative HMM fitting, spectral extraction of latent HMM features, and discriminative classification with SVMs.