Learnijoy
ResearchAI ResearchAI Engineering & DevTools

LLMs Aid Causal Discovery for Adverse Pregnancy Outcomes

Kavimayil P. Komarasamy, Saurabh Mathur, Ameet Soni, David M. Haas, Kristian Kersting, Sriraam Natarajan· August 24, 2026 View original

Key takeaways

  • A neurosymbolic framework combines LLMs and empirical data for causal discovery in APOs.
  • The LLM adaptively proposes hypotheses, which are scored against real-world data.
  • The method recovered all expert-validated causal edges and identified new plausible relations.
  • It offers a powerful approach for causal inference in data-scarce, complex domains.

Who benefits

HealthcarePharmaceuticalsMedical ResearchAI/ML Development

Summary

Researchers developed a neurosymbolic framework that combines LLM knowledge with empirical data scoring to generate plausible causal hypotheses for Adverse Pregnancy Outcomes (APOs). This iterative method adapts LLM proposals based on high-scoring graphs, recovering expert-validated edges and identifying new potential causal relations.

Understanding the causes of Adverse Pregnancy Outcomes (APOs) like preterm birth and gestational diabetes is crucial but challenging due to limited data and incomplete domain knowledge. Traditional data-driven methods often fall short, and Large Language Model (LLM) outputs can be inconsistent. A new neurosymbolic framework has been introduced to address this, aiming to generate more plausible causal hypotheses. This framework iteratively combines the extensive prior knowledge of LLMs with empirical scoring against real-world data. The LLM acts as an adaptive proposal distribution, generating potential causal graphs. These graphs are then scored based on how well they fit the empirical data. High-scoring graphs are used to update the LLM's context, effectively steering subsequent hypothesis generations towards more promising areas of the causal space. Evaluated on a clinical dataset for APOs and their risk factors, the method successfully recovered all causal edges previously validated by experts. Furthermore, it identified additional plausible causal relations that experts had not previously listed, potentially offering new insights for targeted medical interventions. This approach demonstrates a powerful way to leverage LLMs for causal discovery in data-scarce, complex domains.

Why it matters

This framework offers a novel and effective way to uncover complex causal relationships in critical medical domains, potentially leading to better understanding, prediction, and intervention strategies for adverse health outcomes.

How to implement this in your domain

  1. 1Explore applying this neurosymbolic framework to other medical or biological domains where causal discovery is challenging due to data scarcity.
  2. 2Collaborate with medical experts to validate newly identified causal relationships and assess their clinical significance.
  3. 3Develop tools or platforms that integrate LLM-driven hypothesis generation with empirical data scoring for causal inference.
  4. 4Investigate the framework's adaptability to different types of data, such as electronic health records or genomic data, for broader application.

Original post by Kavimayil P. Komarasamy, Saurabh Mathur, Ameet Soni, David M. Haas, Kristian Kersting, Sriraam Natarajan

"arXiv:2608.21079v1 Announce Type: new Abstract: Adverse Pregnancy Outcomes (APOs) such as preterm birth and gestational diabetes can have long-term consequences for both the mother and child, yet an understanding of their causes remains elusive. Causal discovery in this domain is…"

View on X

Originally posted by Kavimayil P. Komarasamy, Saurabh Mathur, Ameet Soni, David M. Haas, Kristian Kersting, Sriraam Natarajan on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Harmony Improves Protein-Ligand Flexible Docking with Torsional Diffusion

Researchers introduce Harmony, a harmonic torsional diffusion framework for flexible protein-ligand docking that explicitly accounts for the periodic geometry of angular variables. This method improves ligand pose accuracy and pocket all-atom reconstruction on benchmarks like PDBBind and enhances the physical validity of generated complexes on PoseBusters.

Maksim Zhdanov, Pavel Strashnov, Vladislav KurenkovAug 24, 2026
AI Engineering & DevToolsAI Research

Multilingual Verifier Bias Impacts RLVR in LLM Mathematical Reasoning

A study reveals that exact-match verifiers in Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Models (LLMs) exhibit significant language-dependent false-negative reward noise in multilingual mathematical reasoning. This bias, particularly pronounced in Japanese, stems from format and script variations, highlighting a cross-lingual selection bottleneck that impedes effective multilingual LLM training.

Chenyu Zhou, Qiliang Jiang, Xu ZhouAug 24, 2026
AI Engineering & DevToolsAI Research

TriPLU Improves Tiny Language Model Performance with Trilinear Product FFNs

Researchers introduce TriPLU, a Trilinear Product Linear Unit, which replaces gated FFNs in tiny decoder-only language models with a direct degree-3 product branch. This approach achieves better validation loss on character-level TinyStories and lower bits per byte on other datasets under low-learning-rate settings, suggesting benefits for small models in specific low-compute regimes.

He ZhangAug 24, 2026