Tessa Green
@tessallator.bsky.social
computational immunology and transcriptomics @KiraGenBio| formerly @MIT@harvardmed and @hgsuuaw healthcare | she/her | 🏳️🌈👩🏻💻
tessadgreen.github.io
tessadgreen.github.io
oh look it’s me (and my favorite dog)
December 17, 2024 at 4:31 PM
oh look it’s me (and my favorite dog)
Some beloved metrics like Spearman and Wasserstein performed relatively poorly. E-distance was comparable to MSE, even though here we directly computed on HVGs rather than in PCA space as in scPerturb—but MSE is faster.
December 17, 2024 at 3:46 PM
Some beloved metrics like Spearman and Wasserstein performed relatively poorly. E-distance was comparable to MSE, even though here we directly computed on HVGs rather than in PCA space as in scPerturb—but MSE is faster.
When applicable, we define the BioRep score as how well the distances correlate with known covariates of each dataset: developmental time, number of CRISPR guides, drug dose. Good metrics increase when cells are more perturbed from control.
December 17, 2024 at 3:46 PM
When applicable, we define the BioRep score as how well the distances correlate with known covariates of each dataset: developmental time, number of CRISPR guides, drug dose. Good metrics increase when cells are more perturbed from control.
A good metric is one where these should-be-identical groups are closer to each other than they are to other, non-control cells (CRP); and where this holds for all five random subsamplings of the control (robustness).
December 17, 2024 at 3:46 PM
A good metric is one where these should-be-identical groups are closer to each other than they are to other, non-control cells (CRP); and where this holds for all five random subsamplings of the control (robustness).
We define three benchmarks, then evaluate them using nine datasets. Two of these (control rank percentile and robustness) we derive by randomly splitting control cells, then seeing how consistently close those are to each other.
December 17, 2024 at 3:46 PM
We define three benchmarks, then evaluate them using nine datasets. Two of these (control rank percentile and robustness) we derive by randomly splitting control cells, then seeing how consistently close those are to each other.
Evaluation metric choice has a big impact on what models we find perform best. So how do we evaluate evaluation metrics for predicting single-cell gene expression? In this preprint, we create a benchmark for population comparison. Tl;dr: use mean squared error
December 17, 2024 at 3:46 PM
Evaluation metric choice has a big impact on what models we find perform best. So how do we evaluate evaluation metrics for predicting single-cell gene expression? In this preprint, we create a benchmark for population comparison. Tl;dr: use mean squared error
Interestingly when I search 'symptoms of T cell exhaustion' (the part that honestly doesn't really make sense) I get a dramatically more cogent response
November 12, 2024 at 6:40 PM
Interestingly when I search 'symptoms of T cell exhaustion' (the part that honestly doesn't really make sense) I get a dramatically more cogent response
Fairly far down the list of complaints about Google's search AI, but I feel like the symptoms/causes/effects/treatment is probably not the optimal organization for this particular term.
November 12, 2024 at 6:40 PM
Fairly far down the list of complaints about Google's search AI, but I feel like the symptoms/causes/effects/treatment is probably not the optimal organization for this particular term.
Very grateful to be at a company that encourages WFH while sick, and for a building with excellent air circulation. I’d feel a lot more comfortable if the other groups in our coworking space did the same, though.
November 12, 2024 at 4:39 PM
Very grateful to be at a company that encourages WFH while sick, and for a building with excellent air circulation. I’d feel a lot more comfortable if the other groups in our coworking space did the same, though.
“Get over here and take my picture, baby, I’m money laundering!”
November 30, 2023 at 6:20 PM
“Get over here and take my picture, baby, I’m money laundering!”
I am personally offended by this graphic from CCC. If you’re doing a Venn diagram the candidates with multiple affiliations need to have their names in the overlap! Sobrinho-Wheeler is in listed separately in four different circles!
November 5, 2023 at 2:51 PM
I am personally offended by this graphic from CCC. If you’re doing a Venn diagram the candidates with multiple affiliations need to have their names in the overlap! Sobrinho-Wheeler is in listed separately in four different circles!
Selfie with a giant Siamese carp—close but not quite!
October 13, 2023 at 7:43 PM
Selfie with a giant Siamese carp—close but not quite!
Success! Manifold earrings definitely worth the re-piercing
October 13, 2023 at 1:30 AM
Success! Manifold earrings definitely worth the re-piercing
Finalizing a paper for publication—-are there actually people who prefer this to a zip of csv files?
October 2, 2023 at 11:38 AM
Finalizing a paper for publication—-are there actually people who prefer this to a zip of csv files?
Pre-defense good luck flowers from my mom, who is expecting to miss the main event because it turns out it’s possible to get COVID twice in three weeks! What an exciting time to be alive!
September 30, 2023 at 7:17 PM
Pre-defense good luck flowers from my mom, who is expecting to miss the main event because it turns out it’s possible to get COVID twice in three weeks! What an exciting time to be alive!
Further update: bright purple successfully mellowed, hopefully won’t crash with my earrings on Monday
September 30, 2023 at 3:29 AM
Further update: bright purple successfully mellowed, hopefully won’t crash with my earrings on Monday
Drop something blue from your gallery
September 8, 2023 at 4:26 PM
Drop something blue from your gallery