TransMeta

TransMeta assembles RNA-seq reads from multiple samples into full-length transcripts to reconstruct consensus transcriptomes and generate unified meta-annotations in GTF format.

Key Features:

• Vector-weighted splicing graph: Represents nodes and edges with vectors whose elements correspond to weights from each sample k.
• Unified meta-annotations (GTF): Produces a consensus meta-annotation in GTF format for all samples in an experiment.
• Individual-sample transcript reconstruction: Generates individual transcript sets for each sample alongside the consensus annotation.
• Cosine-similarity-based combing: Integrates sample information using a cosine-similarity-based combing strategy.
• Label-setting path-searching algorithm: Selects transcript paths using a newly designed label-setting path-searching algorithm.
• Input/output formats: Accepts lists of alignment files in BAM format and outputs assembled candidate transcripts in GTF format.
• Benchmarking performance: Evaluated on simulated and real datasets and reported superior precision and recall at the meta-assembly level versus PsiCLASS, StringTie2 (merge mode), and Scallop combined with TACO.
• Robustness and accuracy: Demonstrated robustness and improved performance across varying coverage thresholds and in individual-sample analyses.

Scientific Applications:

• Multi-sample transcriptome assembly: Reconstruction of consensus transcriptomes from multi-sample RNA-seq experiments.
• Meta-annotation generation: Creation of unified GTF annotations for downstream transcriptomic analyses.
• Per-sample transcript recovery: Extraction of per-sample transcript sets within a combined multi-sample assembly framework.
• Method comparison and benchmarking: Comparative evaluation of assembler precision and recall across datasets and coverage thresholds.

Methodology:

Constructs a vector-weighted splicing graph with sample-specific weight vectors on nodes and edges, applies a cosine-similarity-based combing strategy and a label-setting path-searching algorithm to select transcript paths, takes lists of BAM alignment files as input, and outputs assembled candidate transcripts in GTF format; validated on simulated and real datasets.