xTrimoGene: An Efficient and Scalable Representation Learner for Single-Cell RNA-Seq Data

Authors: Jing Gong, Minsheng Hao, Xingyi Cheng, Xin Zeng, Chiming Liu, Jianzhu Ma, Xuegong Zhang, Taifeng Wang, Le Song

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments also show that the performance of x Trimo Gene improves as we scale up the model sizes, and it also leads to SOTA performance over various downstream tasks, such as cell type annotation, perturb-seq effect prediction, and drug combination prediction.
Researcher Affiliation Collaboration Bio Map Research 2 Tsinghua University 3 Mohamed bin Zayed University of Artificial Intelligence
Pseudocode No The paper includes Figure 1 which illustrates the x Trimo Gene Framework, but this is a conceptual diagram and not structured pseudocode or an algorithm block.
Open Source Code No x Trimo Gene model is now available for use as a service via the following link: https://api.biomap.com/xTrimoGene/apply. This indicates the model is available as an API service, but does not provide access to its source code.
Open Datasets Yes We evaluated x Trimo Gene s performance on cell type annotation task with Zheng68K [39] and Segerstolpe [31] dataset, which has been widely benchmarked.
Dataset Splits No The paper mentions using benchmarked datasets and referring to Appendix 1 for data description and Appendix Table 2 for hyperparameter settings, but it does not explicitly provide specific training, validation, or test split percentages or sample counts in the main text.
Hardware Specification Yes The memory consumption for inference with the x Trimo Gene-100M model is approximately 50GB, whose hardware requirement (Nvidia A100 80G GPU) is beyond some academic labs...
Software Dependencies No The paper does not explicitly list any software components with their specific version numbers (e.g., Python 3.x, PyTorch 1.x, CUDA 11.x) that would be needed to replicate the experiments.
Experiment Setup Yes To test the scale-up ability of x Trimo Gene, we pre-trained three models across multiple compute regions and scales (e.g., from 3M to 100M parameters). The detailed hyperparameter setting is displayed in the App. Table 2.