-
A new LINCS submission experienced timeouts in prod and also shows very slow ingest timings under test in dev environment.
The core fact keywords ETL query seems to be the standout query which is r…
-
Looking at the results in `chemical_CPA/simon/plot_sweep_results.ipynb`
**Hypothesis**: The drug disentanglement might be easier for NN based embeddings as these are chemically motivated.
**Poten…
-
- [ ] documenter les responsabilités
- [ ] signatures, règles de collaborations, mentions des sources
- [ ] procédures et politiques
- [ ] documentation
-
Trying to record vocab usage for a submission with large synonym lists, for example a compound in recent LINCS submissions, hits an implementation limit and fails.
To remove this limit, we need to …
-
ghost updated
2 years ago
-
As it stands there's no parameter finetuning in #85 or #84.
1. Parameters to be tuned (all models):
- `autoencoder_lr`, `autoencoder_wd`
- `batch_size`
- All adversary parameters. Since we crea…
-
- Gene-gene interactions (GGIs) from mRNA co-expression:
GGIs are computed from thousands of randomly selected RNA-seq samples processed uniformly with the ARCHS4 pipeline. The top 10 genes …
-
The Vanilla Model with ~6M parameters is outperforming most of our other model, however these other models have maximum ~2M parameters.
It might be worth scheduling some runs with very large models…
-
I submitted the GTEx data as GTEx. Then I submitted exactly the same data but as LINCS. The question I was trying to answer is “What breaks if someone who is a submitter at two DCCs submits with the w…
-
Generate 3 new Trapnell splits, one for each of the 3 pathways (epigenetic, cell cycle, tyrosine kinase).
In each split, 10 drugs are left out.
For LINCS we need just one new split, where all 30 …