Thread by @stephensammut

1/14 🆕 🚨Accelerated article on #breastcancer just out in @nature! #bcsm

We asked:
1️⃣ What drives response to #chemotherapy in #breast #cancer❓
2️⃣ Can we use #machinelearning to predict response to chemotherapy❓

Link: nature.com/articles/s4158…

A 🧵👇👇👇 @OncoAlert

2/14 We are unable to #predict #response to treatment in clinic…

✅Good response ➡️treatment very effective ➡️better survival
❌Poorer response ➡️treatment less effective ➡️worse survival

Being able to forecast response would be a landmark advance!

3/14 We ran a study in women with #breastcancer @CUH_NHS+@CRUK_CI:
1️⃣who had a #cancer biopsy (which we analysed)
2️⃣received 18 weeks of chemo+/-targeted therapy
3️⃣had surgery (and we measured how much cancer remained)

We associated PRE-therapy profiles↔️POST-therapy response👇

4/14 We measured amount of cancer remaining post-therapy using RCB classification (developed at @MDAndersonNews)

4️⃣categories

No cancer:
1 pCR (complete response)

Cancer found:
2 Minimal: RCB-I
3 Moderate: RCB-II
4 Extensive: RCB-III

⬆️RCB= therapy⬇️effective= ⬇️survival👇

5/14 What did we find?

The baseline pre-therapy features are #monotonically #associated with response! 🤯🤯

𝗠𝗼𝗻𝗼𝘁𝗼𝗻𝗶𝗰 ❓What is that❓

The more (or less) of a characteristic the cancer has BEFORE treatment the less the amount of cancer remaining AFTER treatment 👇👇

6/14 🧬 Pre-therapy #DNA features associated with response:

💥💥Genomic instability!💥💥
- mutations in cancer driver genes (eg TP53, PIK3CA)
- tumour mutation and neoantigen burden
- degree of chromosomal instability
- homologous recombination deficiency
- MHC class I LOH

7/14 The #RNA expression landscape also monotonically associated with response!

Predominantly fashioned by:
- #Estrogen signalling
- Tumour proliferation
- #Immune #microenvironment composition+activity

Look at interplay between the tumour+immune system in these density plots👇

8/14 🔬⚙️ With @fangyingww and @AliDariush2 we computed the lymphocytic density in the tumour H&E #pathology slides using #machinelearning

Relationship between lymphocyte density and response also monotonic‼️

9/14 With @mireiacrispin we combined #clinical+#DNA+#RNA+#digitalpathology features in an #ensemble #machinelearning model 💻 to predict response.

1️⃣ AUC pCR validation dataset = 0.87
2️⃣ Also PRE-therapy predictor score associated with the RCB score POST-therapy! 🤯🤯

10/14 Hang on. Have we just shown that a #machinelearning framework that integrates multi-omic data from the #tumour + its #ecosystem can predict response to treatment before we start it? 🔮

Yes 🙂

We can now predict who will respond really well and who will respond less well

11/14 Why is this #CancerResearch important?

Maybe it’s time to change the way we do #clinicaltrials?💊💉

Predicted response to standard treatment:
✅Good: administer standard treatment‼️
❌Poor: consider recruitment to #clinicaltrials

We have the #technology! @OncoAlert

12/14 We are indebted to so many who have contributed so much:
1. The women who selflessly participated in this study. #Thankyou.
2. Our funders: @wellcometrust @CR_UK @TheMarkFdn
3. Our colleagues at @MRC_BSU @Cambridge_Uni @CRUK_CI @EdinCRC @warwickuni @PeterMacCC

13/14 To our mentors for their unconditional support and invaluable contribution to this work: @CaldasLab @SarahJ_Dawson @OMRueda @SuetFeungChin @paulpharoah @markowetzlab

14/14 💥💥Spoiler alert: Carlos Caldas will present this work at @SABCSSanAntonio #SABCS21 @SusanGKomen Brinker award lecture! 💥💥