Skip to content

v1.7.8

Latest
Latest

Choose a tag to compare

View all tags
@Starlitnightly Starlitnightly released this 03 Nov 23:53
· 211 commits to master since this release
v1.7.8
ad3d331

v 1.7.8

Implemented lazy loading system that reduces import omicverse time by 40% (from ~7.8s to ~4.7s).
Added GPU-accelerated PCA support for Apple Silicon (MLX) and CUDA (TorchDR) devices.
Introduced Smart Agent System with natural language processing for 50+ AI models from 8 providers.
Added and fixed the anndata-rs to support million size's datasets (#336)

PP Module

  • Added GPU-accelerated PCA in ov.pp.pca() with MLX support for Apple Silicon MPS devices
  • Added TorchDR-based PCA acceleration in ov.pp.pca() for NVIDIA CUDA devices
  • Added smart device detection and automatic backend selection in init_pca() and pca() functions
  • Added graceful fallback to CPU implementation when GPU acceleration fails
  • Added enhanced verbose output with device selection information and emoji indicators
  • Added optimal component determination based on variance contribution thresholds in init_pca()
  • Added GPU-accelerated SUDE dimensionality reduction in ov.pp.sude() with MLX/CUDA support
  • Optimize the ov.pp.qc and added ribosome and hb-genes to know more information of data quantity.

Datasets Module

  • Complete elimination of scanpy dependencies for faster loading
  • Added dynamo-style dataset framework with comprehensive collection
  • Added robust download system with progress tracking and caching
  • Added enhanced mock data generation with realistic structure
  • Added support for h5ad, loom, xlsx, and compressed formats

Agent Module

  • Added multi-provider LLM support (OpenAI, Anthropic, Google, DeepSeek, Qwen, Moonshot, Grok, Zhipu AI)
  • Added natural language processing for both English and Chinese
  • Added code generation architecture with local execution
  • Added function registry system with multi-language aliases
  • Added smart API key management and provider-specific configuration

Bulk Module

  • Added BayesPrime and Scaden to deconvoluted Bulk RNA-seq's celltype proportion.
  • Added alignment to alignment the fastq to counts.

Single Module

  • Added ov.single.Annotation and ov.single.AnnotationRef to annotate the cell type automatically.
  • Added ov.alignment.single to alignment the scRNA-seq to counts directly.
Assets 4
Loading