Release OpenSpliceAI v0.0.7 · Kuanhao-Chao/OpenSpliceAI

OpenSpliceAI v0.0.7

variant --batch-size / -b — batched inference on PyTorch models for large speedups on many-variant VCFs (buffers records and deduplicates the shared reference window across a position's alternate alleles). The default (1) preserves the exact original per-variant path bit-for-bit. Tunables: OSAI_CUDNN_BENCH (cuDNN autotune) and OSAI_TF32 (A100 TF32 fast path; set OSAI_TF32=0 for bit-reproducible output).
predict --gene-flank — with -a/--annotation, include real genomic flanking sequence on each side of every extracted gene so the model sees true context instead of N padding at gene boundaries (default -1 → flanking_size/2; 0 = legacy bare-gene body). predict now also warns on sequences shorter than the model's required context and clarifies strand handling when no annotation is given (closes #16).

Now installable from Bioconda: conda install -c conda-forge -c bioconda openspliceai.
Corrected the conda recipe to GPL-3.0 built from the PyPI source tarball + checksum, added GPLv3 license metadata and trove classifiers to setup.py, and pruned three unused dependencies (torchaudio, torchvision, matplotlib-inline).

variant: the reference one-hot encoder now folds every non-ACGT base (N, IUPAC ambiguity codes, gaps) to the all-zero row per its documented contract. ACGTN reference sequence is encoded bit-identically, so real-genome scores are unchanged.
predict: fixed an AttributeError crash in the neg_strands reverse-complement path of get_sequences.

Expanded the pytest suite to ~300 tests at ~96% line coverage of the packaged pipeline, with characterization tests locking the cross-subcommand hyperparameter table, encode↔decode round-trips, and batched-vs-sequential variant equivalence. Added a one-command reproducible test entrypoint (Makefile) with a coverage-floor gate.

Install: pip install openspliceai==0.0.7 · PyPI: https://pypi.org/project/openspliceai/0.0.7/ · Docs: https://ccb.jhu.edu/openspliceai/