A note from our CPO
Closer to the clinical truth
I run product at Biobase. We build a semantic layer over native clinical data. Practically, this means we are making the raw clinical data more accessible to consumers, namely AI agents. Our products map data to shared standards so pharma, drug developers and AI teams can work closer to the clinical truth. Currently, we are investing in AI agents for trial feasibility and patient matching, grounded in real, geographically linked clinical data. Designing a trial? Book time with me.
Chief Product Officer, Biobase
The Biobase Brief, issue 02 · October 2026. Adapted for the web from the Life Sciences edition.
RNA and protein can tell different stories
A proof-of-concept study matched individual cells from colorectal liver metastases across sequencing and imaging platforms. RNA and protein levels for PD-L1, TIM-3, and IDO correlated only weakly. In one region, PD-L1 protein was observed in roughly half the cells while its RNA was detected in fewer than 5%.
The small sample and single-tumor setting limit generalization. The finding supports asking whether a research question needs paired measurements; it is not a validated clinical rule about these markers.
Eapen et al., Cancer Research Communications 6, 1963 (2026).
Chromatin information from older tissue
A method for profiling single-cell chromatin accessibility was applied to FFPE lymph node samples stored for 8–12 years. It recovered more than 12,000 high-quality cells across five cell types. Analyses of paired primary and relapsed tumors also recovered signals distinguishing the two states from material stored for 6–13 years.
These results broaden the research questions that archival material may support. They do not guarantee assay suitability for every old block.
Yadav et al., Nature Communications 16, 10022 (2025).
Whole-genome analysis needs to account for FFPE artifacts
Across 56 matched fresh-frozen and FFPE pairs, FFPE processing produced a median 20-fold increase in false mutation calls. The FFPErase filter improved small-variant precision from 52% to 92% at 80% sensitivity. In a comparison with an FDA-approved panel in 34 patients, it retained 99% of the panel’s findings and identified additional clinically relevant events in 14 patients, mainly structural variants.
The authors note remaining copy-number noise and the need for retraining before applying the approach to exome or panel data. These are research performance results, not a claim of clinical approval for the filter.
Domenico et al., Nature Communications 16, 10649 (2025).
Ask for the relationship between measurements
For sourcing, specify which observations must come from the same patient, block, section, or region. “Matched data” can describe very different relationships. State which one your analysis requires and how missing measurements should be handled.
Use the biospecimen request guide to bring cohort, material, and linked-data requirements into one brief.