TENEX¶
TENEX (TENET eXtremely optimized) computes pairwise transfer entropy (TE) on the GPU to infer gene regulatory networks (GRNs) from single-cell RNA-seq data. It is a high-performance reimplementation of FastTENET, achieving up to a 2,203x speedup while preserving numerical accuracy.
Why TENEX¶
Transfer entropy quantifies directed information flow between genes along pseudotime. Estimating it for every ordered gene pair requires counting a 3-D joint histogram over discretized expression values, and this counting is the bottleneck. TENEX accelerates it by
- packing each histogram triplet into a single integer address and counting directly on the GPU, replacing the sort-based counting of FastTENET, and
- selecting, per dataset and device, the fastest of several CUDA kernels.
Performance¶
Matched single-GPU comparison on NVIDIA PRO 6000 Blackwell (median of 3 runs).
| Dataset | Genes | FastTENET | TENEX | Speedup |
|---|---|---|---|---|
| mESC | 3,281 | 33.28 s | 0.334 s | 100x |
| Skin | 1,960 | 115.76 s | 0.216 s | 536x |
| Zebrafish | 25,258 | 18.06 h | 42.92 s | 1,515x |
| CeNGEN | 22,469 | 52.99 h | 86.60 s | 2,203x |
Across all datasets TENEX reproduces FastTENET TE values within float32
precision when no bin coarsening is applied, with Pearson correlation
r >= 0.9999.
Features¶
- Adaptive kernel selection: automatic selection of the best CUDA kernel for the data and hardware characteristics.
- Multi-GPU support: thread-based parallelism with no spawn overhead.
- GPU-native preprocessing: discretization and bin remapping entirely on GPU.
- Multiple inference methods: FDR, CLR, Network Deconvolution, and a surrogate-based statistical test.
Where to next¶
License¶
TENEX is released under the TENEX Non-Commercial License. Use is permitted for non-commercial purposes only, and the license must be reproduced in copies and derivative works. For commercial licensing, contact Daewon Lee (dwlee@cau.ac.kr).