Eric Kernfeld

There is a lot of new work coming out relevant to predicting transcriptomic responses to perturbation.

The problem with gene-wise characterization of genetic interventions When you target a gene with CRISPRi, or something, its transcript goes down (or up).

Bias-variance tradeoffs for virtual cell evals are weird Consider a Perturb-seq experiment with controls $Y_1 … Y_C$, perturbation G applied to samples $Z_1 … Z_C$, and other perturbations applied to samples $X_1 … X_N$, with N»C.

The comp bio job market is scarier than usual: 2023 was a biotech bankruptcy bloodbath, 2024 was a little better, and as far as layoffs, 2025 has been worse.

In October 2024, I twote that “something is deeply wrong” with what we now call virtual cell models.

Stratified linkage disequilibrium score regression (S-LDSC) is a workhorse technique at the intersection of human genetics and functional genomics, but its exposition is heavily genetics-coded.

Stratified linkage disequilibrium score regression (S-LDSC) is a workhorse technique at the intersection of human genetics and functional genomics, but its exposition is heavily genetics-coded.

Stratified linkage disequilibrium score regression (S-LDSC) is a workhorse technique at the intersection of human genetics and functional genomics, but its exposition is heavily genetics-coded.

Stratified linkage disequilibrium score regression (S-LDSC) is a workhorse technique at the intersection of human genetics and functional genomics, but its exposition is heavily genetics-coded.

Linkage disequilibrium score regression (LDSC) is a workhorse technique at the intersection of quantitative genetics and functional genomics.