Causal Inference Sub Group 2025

AWG Projects AIML AWG Open Projects
1

Dear all,

Thanks a lot for showing active interest in this subgroup & responding the earlier survey ( I have got response from 20 people which is super exciting ) .

I am creating this topic on the portal to have one active page for this subgroup.Many of you were interested in understanding more details about work done in this subgroup and plans for the year 2025. I have summarized the same below, Feel free to post your questions/doubts related with this subgroup. I will also be creating an event for our regular meeting cadence.

Background

The causal Inference subgroup is one of the subgroups in AI/ML AWG ( analysis working group) under the Open Science Data Repository AWG Community ‘Space.’

This subgroup works on building and improving solutions for causal inference in existing and new use cases of biological numerical data.

Currently, CRISP ( AI4LS/crispv1.1 at main ¡ nasa/AI4LS ¡ GitHub ) is used for causal Inference. It is an ensemble model and primarily designed for structured datasets. The reference paper associated with this is available at Prototyping CRISP: A Causal Relation and Inference Search Platform applied to Colorectal Cancer Data | IEEE Conference Publication | IEEE Xplore

Roadmap 2025

The three primary objectives for this year are -

  1. Improving the accuracy of the existing CRISP model & testing/validating new ML models can help improve the coverage/accuracy. One approach to this is to find a biomedical dataset(s) where the causal relationships have been already established in the lab, and validate our causal algorithms on that dataset. Another approach is to use a structural equation modeling (SEM) approach in which a known causal structure is embedded in a dataset.

  2. Evaluating other machine learning models & approaches in causal inference that can also help in unstructured data ( text & images) ,We have a team working on a “multi-modal CRISP”, version of CRISP that can take numeric matrix data and images. It would be great for this subgroup to evaluate and ultimately use that version of CRISP to answer some scientific questions

  3. Do a research project to answer a key scientific question in spaceflight, using the improved causal inference algorithms

@lauren.sanders @rtscott2001 @james.casaletto

4 Likes
2

Thank you @pmisra30 ! I am looking forward to our first meeting on Jan 15!

2 Likes
3

So excited to hear about this. Cheers and thanks Pramod! :tada:

1 Like
4

Thanks for the update @pmisra30. Project sounds very interesting, see you at the meeting!

2 Likes
5

Could anyone point us to datasets similar to the multi-omic colorectal cancer dataset mentioned in the CRISP study, or other biomedical datasets with causal structures suitable for validation?

1 Like
6

@james.casaletto - able to answer abdelghani’s question? @abdelghani.belgaid @lauren.sanders

1 Like