Heng Pan

Computational Biologist @WeillCornell

[Dan Landau @landau_lab]

Finally, although we built MethSig for NGS data, and see it as the future of this field, MethSig also adds value to array based datasets.

[Dan Landau @landau_lab]

Prognostic models based on DNA methylation drivers separated high and low risk CLL, including in an independent validation cohort, and after accounting for other known risk factors.

[Dan Landau @landau_lab]

We validated top CLL TSG hits through CRISPR/Cas9 KOs, showing that they result in increased fitness in unperturbed growth as well as preferential fitness with leading agents.

[Dan Landau @landau_lab]

We also devised metrics of success, for example likely DNA methylation drivers are expected to associate with gene silencing and clinical outcomes. This analysis shows a radical increase in performance in ROCs.

[Dan Landau @landau_lab]

Indeed, like in cancer genomics, we know that DNA methylation changes occur at different rates across the genome. Applying models with a uniform background result in an unacceptable rate of false positives....

[Dan Landau @landau_lab]

@dragon_heng and friends designed MethSig, a statistical inference model that accounts for this varying background rate, resulting in better calibrated QQ plots compared with benchmarks and higher reproducibility across cohorts.

[Dan Landau @landau_lab]

Excited to share new work @CD_AACR!
https://t.co/V3rUx55mGO
Sophisticated background models in cancer genomics distinguish driver from passenger mutations. As most cancer DNA methylation changes are stochastic can we likewise create better models to identify driver events?
🪡 /n