top of page

Discovering potent and specific kinase inhibitors with the DeepMirror App

Kinases are frequently dysregulated in disease and are an interesting yet challenging target against which to develop drugs. In this blogpost, we use the DeepMirror App to identify novel, potent, and selective kinase inhibitors.


Kinases are proteins that activate or deactivate other proteins to regulate a variety of cellular functions. Mutations or dysregulations can lead to severe diseases. For example, overexpression of kinases may drive uncontrolled cell growth and cancer (e.g., BCR-ABL in Chronic Myeloid Leukaemia), and in diabetes, kinases like PKC contribute to insulin resistance.


There are more than 500 kinase genes in the human genome, and different kinase families often share many structural similarities. These similarities make it challenging to design small molecules that discriminate between different kinases. Additionally, kinases interact in complex networks with other proteins, making it challenging to modulate their effects on specific proteins. Balancing potency and specificity of kinase inhibitors requires a deep understanding of kinase structure, dynamics, and the underlying biology, in combination with sophisticated design strategies and rigorous experimental validation.


Can AI help generate compounds that bind one kinase but no related ones? In this blogpost, we used the DeepMirror App to test how experimental data on ligand-kinase binding against related kinases can be utilised to inform molecular design decisions.


DDR1 and DDR2 as drug targets

DDR1 and DDR2 are cell surface receptors that play crucial roles in cancer progression, making them intriguing drug targets. DDR1 and DDR2 are involved in mediating signals from a cells’ environment, impacting processes such as cell proliferation, invasion, and metastasis. DDR1 is particularly relevant in various cancer types, including breast and lung cancer, where it has been linked to tumour growth and angiogenesis. DDR2, on the other hand, is known to promote cancer cell migration and invasion in cancers like ovarian and pancreatic cancer. To the best of our knowledge, no DDR1/2 inhibitors are currently under clinical investigation, but many small molecules have shown promise in pre-clinical studies as attractive candidates for anti-cancer drug development (Denny and Flanagan, 2021). We chose DDR1 and DDR2 as our targets against which to try and identify novel drug targets using the DeepMirror App.


Specific and potent kinase inhibitors against DDR1 and DDR2

We used a publicly available kinase inhibitor dataset that contained activity values (pIC50) against DDR1/2 and many other kinases (Laufkötter et al., 2020). To test whether the DeepMirror App could help us find a molecule that could be active against DDR1/2, but not against other kinases, we curated this dataset by filtering only related kinases known to commonly cross-react with DDR1/2 inhibitors (Elkamhawy et al., 2021) . These included c-Src, B-Raf, BTK, VEGFR, BCR, p38, Abl, Raf-1 etc.


In total, we had a dataset of 4746 small molecule inhibitors and 12 kinases. The dataset was missing many activity values, as inhibitors were rarely tested against all kinases.

Screenshot of the Deepmirror App upload with kinase inhibitor dataset

1. We first uploaded the curated kinase inhibitor dataset to the DeepMirror App and generated 1,000 compounds using the Generate with AI feature.

2. Next, we predicted activity/pIC50 against all kinases for all compounds. The DeepMirror App learned the relationship between the molecular structures of the compounds and the activity against the target kinases, to fill the missing values of pIC50 in the dataset.

Screenshot of the Deepmirror App generated kinase inhibitor compounds and their predicted activity

3. Finally, we filtered for novel compounds with a moderate to high potency against DDR1/2, and a low to moderate potency against all other off-target kinases to identify potent and specific inhibitors of DDR1/2.

Screenshot of the Deepmirror App filtering novel compounds for specificity and potency

After applying these filters, 5 AI generated compounds remained. Of those, 4 had good synthetic accessibility, i.e. were predicted to be easy to synthesise.


In 3 simple steps, we conducted an in silico screen of novel, AI-generated compounds using experimental data.



Get in touch at hello@deepmirror.ai with comments or ideas or if you want to try the DeepMirror App yourself.





References

Denny, W.A., Flanagan, J.U., 2021. Inhibitors of Discoidin Domain Receptor (DDR) Kinases for Cancer and Inflammation. https://doi.org/10.3390/biom11111671

Elkamhawy, A., Lu, Q., Nada, H., Woo, J., Quan, G., Lee, K., 2021. The Journey of DDR1 and DDR2 Kinase Inhibitors as Rising Stars in the Fight Against Cancer. Int J Mol Sci 22. https://doi.org/10.3390/IJMS22126535

Laufkötter, O., Laufer, S., Bajorath, J., 2020. Identifying representative kinases for inhibitor evaluation via systematic analysis of compound-based target relationships. Eur J Med Chem 204, 112641. https://doi.org/10.1016/J.EJMECH.2020.112641

 

226 views0 comments

Related Posts

See All

コメント


Want to stay up to date with DeepMirror news?

Thanks for subscribing!

bottom of page