Discovery of Highly Active Kynureninases for Cancer Immunotherapy through Protein Language Model

This article has been Reviewed by the following groups

Read the full article See related articles

Listed in

Log in to save this article


Overcoming the immunosuppressive tumor microenvironment is a promising strategy in anticancer therapy. L-kynurenine, a strong immunosuppressive metabolite can be degraded through kynureninases. Through homology searches and protein language models, we identified and then experimentally determined the efficacy of four top-ranked kynureninases. The catalytically most active one nearly doubles turnover number over the prior best, reducing tumor weight by 3.42 times in mouse model comparisons, and thus, presenting substantial therapeutic potential.

Article activity feed

  1. I can see how some of this can be pieced together with the publicly available Colab, but there are references to some in-house python scripts that would be good to make publicly available as well

  2. We’ve developed a user-friendly Jupyter Notebook, accessible via Google Colaboratory, designed for training customized prediction models using protein sequences and experimental data provided by users in FASTA format.

    I don't see a link to where this is available?

  3. To predict the kinetic parameter kcat/KM for enzymes, we followed the variant prediction workflow proposed in the ESM repository (“examples/sup_variant_prediction.ipynb”).

    I think even if you followed a publicly available notebook, it would be great to have your own code publicly available as well

  4. Our work demonstrated that homology search combined with pLLMs can detect and prioritize highly catalytically active therapeutic enzymes even when only little labelled training data is available

    I really enjoyed the brevity of how the work was communicated! I also enjoyed how this is a case of taking computational predictions, narrowing them down, and making experimental validations/comparisons. I am wondering if your searches could be enhanced through structure-based comparisons/clusters as well as sequence-based? I think most of the hits were bacterial, but I could imagine there are probably other organisms or at least distantly related bacteria that could have enhanced KYNase properties as well.

  5. As a general trend, our predictor predicts higher kcat/KM for sequences from bacteria than from eukaryotes.

    This is interesting and I think readers might want to know more about this observation and the implications - is this expanded upon somewhere else or in the Methods?

  6. We implemented an easy-to-use webserver (see Methods and Data availability) for researchers to conduct similar analyses using their own measures.

    I think highlighting the webserver directly in the text and not just in the Methods section, and maybe also mentioning this development in the abstract would be really useful to readers and bring more visibility to this useful tool!

  7. Trained on an 80/20 split of 159 experimentally measured sequences

    I may have missed something, but is this based on experimental measures that were performed as part of this study or were they already available?