BTW TMB which is used to identify MMRd cases for ICI, has poor specificity, often calling samples with other signatures as TMB high.
Consistent in several cohorts.

Using signatures gives much greater specificity for clinical classification.

11/n

We created a multinomial classifier PRRDetect that can call MMRD, poly dysfunction, combined, or neither.

Performance was better than any other classifier or similar ilk.

10/n

We do highlight an APOBEC indel signature!

Mechanism proposed therein.

8/n

We applied the new system on WGs cancers in 4,775 WGS Genomics England cases, in seven tumor types, identifying 37 signatures.

10 had features reported by COSMIC and we matched these and kept their nomenclature.
We found 27 new signatures. See paper for details.

7/n

Now you can see gene specific differences more easily and this we proved by contrasting the COSMIC system to our 89-channel system using lots of different signature extraction tools from the community 😀

6/n

Our system spreads signal from the two channels in COSMIC to more channels, becoming more informative for discerning biological differences between signatures

5/n

If all indels are aggregated,
all signal is focused on just two channels in COSMIC.

We propose an alternative classification using 5’ and 3’ flanking nucleotides (like subs!), remove cap of 5bp at repeats, amongst others..
resulting in an 89-channel classification system
4/n

But using the current indel classification (COSMIC), we have difficulties “seeing” them.

They get dumped into COSMIC sigs InD1/InD2 (normal sigs of replication slippage) becoz the COSMIC classification aggregates signal at polynucleotide >5bp into one channel

3/n

Mismatch repair deficiency (MMRD) & polymerase dysregulation, collectively referred to as post-replicative repair deficiency (PRRD),
ain’t just one entity.

We created isogenic CRISPR KO models of many PRR genes showing clear gene-specific differences (subs & indels)

2/n