Variational Mixtures of ODEs for Inferring Cellular Gene Expression Dynamics
Authors: Yichen Gu, David T Blaauw, Joshua Welch
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated our method on 6 different sc RNA-seq datasets: pancreatic endocrinogenesis (PE) (Bastidas-Ponce et al., 2019), dentate gyrus (DG1,DG2) (Hochgerner et al., 2018; La Manno et al., 2018), embryonic E18 mouse brain cortex from 10X Genomics (MB1)2, the erythroid lineage from mouse gastrulation (ET) (Pijuan-Sala et al., 2019), and part of a whole mouse brain development dataset (MB2) (La Manno et al., 2021). We used three metrics mean squared error (MSE), mean absolute error (MAE), and log likelihood (LL) to assess how well each method fits the data. For our two models, we calculated these metrics on both a training dataset (70%) and held-out test dataset (30%). |
| Researcher Affiliation | Academia | 1Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, United States 2Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, United States. Correspondence to: David Blaauw <blaauw@umich.edu>, Joshua Welch <welchjd@umich.edu>. |
| Pseudocode | No | No pseudocode or algorithm blocks are explicitly presented or labeled in the paper. |
| Open Source Code | Yes | Source code is available online 1. 1https://github.com/welch-lab/VeloVAE |
| Open Datasets | Yes | We evaluated our method on 6 different sc RNA-seq datasets: pancreatic endocrinogenesis (PE) (Bastidas-Ponce et al., 2019), dentate gyrus (DG1,DG2) (Hochgerner et al., 2018; La Manno et al., 2018), embryonic E18 mouse brain cortex from 10X Genomics (MB1)2, the erythroid lineage from mouse gastrulation (ET) (Pijuan-Sala et al., 2019), and part of a whole mouse brain development dataset (MB2) (La Manno et al., 2021). |
| Dataset Splits | No | The paper mentions training on '70% of the data' and evaluating on a 'held-out test set', but does not explicitly describe a separate validation split or its proportion. |
| Hardware Specification | Yes | As a rough benchmark, we trained our model for 600 epochs with an NVIDIA Tesla V100 GPU and ran sc Velo on a single core of a 2.4 GHz Intel Xeon Gold 6148 CPU. |
| Software Dependencies | No | The paper mentions using the ADAM optimizer and MLPs, but does not provide specific version numbers for software libraries or dependencies like PyTorch, TensorFlow, or Python. |
| Experiment Setup | Yes | For all experiments, we performed minibatch stochastic gradient descent using the ADAM optimizer with learning rate 2 10 4 and batch size of 128. |