OncoNeo400 - A new Precision Oncology Research AI tool on BioAIWorks
The platform is built only for Research and Development purposes

OncoNeo400 - A new Precision Oncology Research AI tool on BioAIWorks

In this edition the OncoNeo400, novel Precision Oncology Research AI tool on BioAIWorks platform (bioaiworks.com). What is the main novelty, being able to predict the immunogenicity of Neoantigens (short tumor specific peptides on tumor cells) and their recognition by various immune system components.

The tool is now in beta v1 version i was developing for some time and will be developed further in Partnership with experienced Oyanalytika team, Gene Garrard Olinger Jr. , Ena Bromley and the supertalented Isak Bromley and Tristan Olinger .

Interested Biotech companies, Oncology Research teams may ask for a demo by contacting me directly or the team above. It should be noted that this product is for Research and Development purposes only (no part of this digital product is to be considered medical advice).


So what do we see as initial impact? Use cases?

  1. Precision Oncology Vaccine Research.
  2. CAR T cells Research and Computational segment of Development
  3. Neoantigen Research and Development

Future directions?

4. We plan to introduce a An AI agent system which will be making large libraries of predictions and make largest ever dataset library of Neoantigen predictions.

5. We will expand and connect Neoantigen prediction systems to Proteomics, Knowledge graphs and evolving AI systems.

Now a bit on how the system is used in this moment...

The user is presented with an option to specialize the AI analysis with various disease types from the Oncology Research field. Further the user may select Precision approach by selection HLA alleles of the immune system (associated with T cells) and finally add a B cell prediction model. It should be noted that multuple models are associated with the final prediction. Models PEP92, B75 and the T800 are associated with MHC I, B cell and T cell response based prediction.

What makes this model characteristic is that its a combination of small super focused models trained on super curated datasets. This means they will achieve high accuracy and speed at the same time. The AI system can make a 10 000 neoantigen predictions in a matter of around 10 seconds.

Once the selection is made the user may opt for Use AI to made predictions, which will start the AI models to predict the immunogenicity of known Neoantigens for that specific type of disease, in the example above Melanoma was selected and various HLA genotypes.


The AI system will make thousands of predictions and present them as the heatmap patterns to the user. Also every prediction is assigned with a specific prediction probability which can be later used for scoring the top immunogenic neoantigens. Note that there is the Construct sequences button which will be used after AI system is initiated.


As it can be seen, the system makes multiple predictions for each Neoantigen peptide, depending on the HLA genotype ( T cell dependent ) and the IgG B cell from multiple AI models. All the predictions per peptide are scored in a special probabilistic scoring system to give every Neoantigen predicted a final predicted immunogenicity score.


The construct Sequences module can not be used to backtrace the best scoring Neoantigens to their proteins of origin, computationally backengineer mRNA and DNA sequences corresponding to it and present the final composite predictions scores.

Significant portion of the Neoantigen information and training data was retrieved from IEDB, special thanks to IEDB for developing the repo.

Note that this is the beta v1 version which is in production, but we are looking to further enhance the capabilities of this AI system. Any feedback is much appreciated and you may provide it the me or the Oya team mentioned at the top of the article.

Those interested in further investing or collaborating with this tech may also request a meeting, demo and a more detailed walkthrough of the technology and the data used to train the models.


Thanks for sharing this! This looks like a very interesting and useful tool to develop. In your article you mention that the AI produces a prediction probability. Do you know if this is this a predictive pvalue or 1 - pvalue (predictive confidence level/margin of error), a Bayesian predictive probability (from a predictive belief distribution), or an estimated proportion/sampling probability (analogous to estimates from a logistic regression)? Thanks again!

Darko Medin

Data Scientist and a Biostatistician. Developer of ML/AI models. Researcher in the fields of Biology and Clinical Research. Helping companies with Digital products, Artificial intelligence, Machine Learning.

21 小时前

Important notice. The AI system itself is made only for Research and Development purposes

赞
回复

要查看或添加评论,请登录

Darko Medin的更多文ç«