I have developed a Multi agent pipeline which has the following flow:
- Identify disease from the user query.
- Identify gene associated with the diseases from publication in PubMed using RAG. (Enhance it to get details from Open Targets, DisGeNET, or GWAS Catalog)
- (Enhancement - Identify protein by querying databases like UniProt, Ensembl, or NCBI Gene using the gene symbol)
- Query ChEMBL find small molecules that show strong potency (IC_50 values) against that protein target and check if it is already FDA approved.
The flow is: Disease -> Gene -> Protein -> Chemical Compound -> Approved Drug
70% of these pipeline is ready, I am working on the enhancement.
Like to get tis reviewed and understand if I am on the right path.
I am using Gemini as LLM and text embedding model provider.
1 answer
Alas, getting something done 70% is the easy part ... fundamentally the problem is that data is not well structured
I think the first thing that needs to be done is fixing the data
Here is a good blogpost from DNA Nexus:
https://blog.dnanexus.com/an-ideal-state-for-bioinformatics-data
Being a bioinformatician is like being a forensic accountant for a chaotic-neutral billionaire who keeps his receipts in a series of wet shoeboxes and occasionally writes key clinical outcomes in crayon on the back of a Sizzler menu.
Log in to answer this question.
It is unclear if you want someone to review just the steps noted above or the actual pipeline itself. Please clarify that. Since no links are provided for the pipeline that is not possible at present.
Thanks GenoMsx for the reply. The pipeline built is in progress and will share it once done. I like someone who is experienced in this field to comment if these kind of pipeline make sense or it is too trivial. I am a computer engineer and new to bioinformatics.
While your pipeline may be more user friendly and integrated, this type of information can be found by searching a couple of sites. Take a look at OMIM (https://omim.org/ ) and ClinPGx (https://www.clinpgx.org/ ).