Editorial & Scientific Analysis
Rigorous analysis on spatial transcriptomics failure modes, single-cell foundation model stress tests, and statistical pitfalls in modern genomics.
A systems biology perspective on the Arc Institute State paper in Cell, examining whether large scale perturbation models capture authentic cell fate bifurcations.
A systems biology perspective on the Arc Institute State paper in Cell, examining whether large scale perturbation models capture authentic cell fate bifurcations.
Commentary on context dependent therapeutic mechanisms and testing intervention predictions beyond training contexts.
Commentary on context dependent therapeutic mechanisms and testing intervention predictions beyond training contexts.
Tahoe-100M's 95.6 million cells cover 379 unique compounds; scale narratives that blur arms with molecules overstate what any model can learn.
Tahoe-100M's 95.6 million cells cover 379 unique compounds; scale narratives that blur arms with molecules overstate what any model can learn.
AI discovered molecules clear Phase I at 80 to 90% against a 40 to 65% historical rate, but Phase II shows no improvement, around 40% versus 37%; AI is strong at molecule properties and has not yet shown it picks the right biology.
AI discovered molecules clear Phase I at 80 to 90% against a 40 to 65% historical rate, but Phase II shows no improvement, around 40% versus 37%; AI is strong at molecule properties and has not yet shown it picks the right biology.
How spatial transcriptomics captures the selective collapse of proximal tubule S3 segments and the emergence of injured states 4 hours post-ischemia.
How spatial transcriptomics captures the selective collapse of proximal tubule S3 segments and the emergence of injured states 4 hours post-ischemia.
Evaluating the trade-offs between 2 million single cells and whole-transcriptome spot deconvolution across 128 kidney spatial samples.
Evaluating the trade-offs between 2 million single cells and whole-transcriptome spot deconvolution across 128 kidney spatial samples.
A systematic re-analysis of 128 publicly available kidney spatial-transcriptomics datasets across five technologies, harmonized and open-sourced via SpatialRenal.
A systematic re-analysis of 128 publicly available kidney spatial-transcriptomics datasets across five technologies, harmonized and open-sourced via SpatialRenal.
Why the low background of CUT&Tag causes ChIP-seq peak callers to misidentify Tn5 accessibility hotspots as genuine transcription factor binding sites.
Why the low background of CUT&Tag causes ChIP-seq peak callers to misidentify Tn5 accessibility hotspots as genuine transcription factor binding sites.
Grillo et al. (Nature Genetics 2026): CRISPRi repression of LTR12C transposable elements forces leukemic stem cells to differentiate — a targetable chromatin state that encodes a cell-fate decision.
Grillo et al. (Nature Genetics 2026): CRISPRi repression of LTR12C transposable elements forces leukemic stem cells to differentiate — a targetable chromatin state that encodes a cell-fate decision.
Ahlmann-Eltze et al. found seven deep learning models fail to beat a simple additive baseline at perturbation prediction; the discipline this creates is worth more than any model.
Ahlmann-Eltze et al. found seven deep learning models fail to beat a simple additive baseline at perturbation prediction; the discipline this creates is worth more than any model.