eGoT Featured in Northeastern Global News
Northeastern Global News reports on eGoT, the enhanced graph-of-thoughts algorithm developed in the Neural Dynamics Lab, describing its approach to multi-domain knowledge retrieval and hallucination reduction in large language models.
On July 30, 2026, Northeastern Global News published a feature on eGoT, the enhanced graph-of-thoughts algorithm developed in the Neural Dynamics Lab. The article, written by Katya Poltorak under the title “Why eGoT gets better answers: Teaching AI to connect the dots across domains,” describes the motivation for the algorithm, its retrieval strategy, and two of the biomedical case studies from the underlying paper, which appeared in Bioinformatics in July 2026 following acceptance at ISMB 2026.
Context
Large language models frequently produce fabricated answers rather than reporting uncertainty, and they integrate poorly across fields of knowledge that are published in separate literatures. In biomedicine, where PubMed alone indexes more than forty million citations, this fragmentation limits reliable synthesis. The article presents eGoT as addressing both failure modes: constraining hallucination while retrieving and connecting evidence that spans domains.
Method Overview
eGoT constructs a graph of concepts relevant to a query. Concepts form nodes that can be expanded into branches, and branches can merge to represent synthesized ideas. For a biomedical question, the algorithm directs the underlying model to scan the available literature and to accumulate sufficient relevant evidence from multiple sources before composing an answer.
Ayan Paul, who leads the lab, is quoted in the article describing the retrieval requirement by analogy: “Imagine you have a bag filled with multicolored balls. To give yourself a fighting chance, you need to grab a big handful.” The design premise is that adequate, targeted evidence is a precondition for sound reasoning, and eGoT’s retrieval procedure is constructed to satisfy it.
Case Studies
The feature summarizes two case studies from the paper. In the first, eGoT traced an indirect pathway by which greenhouse gas emissions affect lupus outcomes through increased ultraviolet radiation exposure, connecting environmental and clinical literatures. In the second, on small cell lung cancer, the algorithm identified associations between transcription factors and clinical features of the disease. Both cases illustrate retrieval across research areas that are ordinarily analyzed in isolation.
Publication and Team
The research was led by Nihar Sanda, who headed the project, with Ayan Paul, Research Associate Professor at the Institute for Experiential AI, and Vito Quaranta, systems biology professor and strategic adviser to the Provost, contributing to the biomedical framing. The paper appeared in Bioinformatics in July 2026.
The full feature is available at Northeastern Global News: Why eGoT gets better answers: Teaching AI to connect the dots across domains.
