Hi, here are some things I’d like to make sure are moving forward. 1. Demonstration of our ability to compute drug descriptors for a large set of candidates in parallel. Ideally we would set up some runs that would compute mordred descriptors for each of the compound in the lists we have in this set of files. Is this something we can get running on Theta? 2. Prototype of our generative model that can be run in parallel to sweep through latent space and generate large stream of SMILEs examples. I’d like to get a version of our VAE models that can run in inference model and do some runs to generate 1M, 10M, 100M example compounds by sampling the latent space in some clever fashion. Ideally I’d like to take the set of generated compounds and search against those in the datasets in #1 to figure out how many are duplicates and how many are novel. For anything novel I would like to compute the Mordred descriptors and store them along with the smiles. Perhaps the right way to do this is to have the inference model directly call mordred so that what comes out of the model is SMILES + Descriptors .. and round robin joined to the RNAseq and SNP features for the Tumors (both CL and PDX) 3. Prototype of our single drug prediction model with UQ in regression mode on growth that is optimized for inference and that can take input from streams or a file. The idea here is that we we will have some number of these models that are running in parallel with #2 With UQ running we want to get out of the model both an estimate of the growth for the given drug/tumor combination but also an estimate of the error on that estimate. This error estimate will need to come from the calibration curves from the UQ. 4. A way to capture the prediction for each generated drug of all the cell lines and PDX models (think of this as a vector of predictions of order ~1K values and error) that we would want to associate with the drug candidate. 5. Some comparison of the throughput of running #1 coupled to #3 vs #2 coupled to #3. Perhaps we can iterate on this via email this week —rick