Technology

AI research putting clinical reasoning at the frontier

The Mendel Approach

Integrating large language modeling and Symbolic AI architectures to address the weakness of each, providing an AI that is capable of clinical reasoning and cognitive modeling.

Mendel is best suited for clinical applications thanks to its advanced reasoning and patient-centric view of the data.

"Triangle + Triangle = Square" icon

Clinical Reasoning

Strong semantic reasoning capabilities and patient-level coherence modules to consolidate a patient journey even when presented data and docs have conflicting facts.
Hallucination icon

Input

Mendel’s architecture grounds the output against a expert-curated knowledge representation of the domain of medicine, avoiding the pitfalls of hallucinations. It is either consistently right or consistently wrong.
Guarded output icon

Guarded Output

Mendel architecture lifts the constraints of context windows and equally works on structured (claims, HL7, etc.) and unstructured (PDFs, notes, Path reports, etc.), avoiding the need to tailor architectures in a statistical-only approach.
Explainability icon

Explainability

Mendel’s technology architecture facilitates instant insight into the failure modes of underlying algorithms. In contrast, incorrect predictions made by a statistical-only approach are more challenging to diagnose.

The Mendel approach surpasses existing LLMs in its advanced reasoning capabilities

In a recent study, we found that solely using large language models (LLMs) like GPT-3.5 decreases Mendel’s system performance by 64.72% on average across 13 medical variables. However, Mendel’s hybrid model combining symbolic clinical reasoning models with LLMs significantly improves interpretation of electronic medical records.
Table of comparing Hypercube, ChatGPT4, and LLAMA on F1, Processing Time, and ~ Cost Per Query.
F1
Processing Time
~ Cost per Query
Hypercube
70.36%
12 milliseconds
$0.01
Chat GPT4
40.54%
4.38 hours
$128
LLAMA-2
34.53%
7.02 hours
$9.30

Hypercube: Breakthrough Hypergraph for medicine.

The clinical domain is high-dimensional, with relationships among data points being complex and not limited to pairs, making knowledge graphs and relational databases unsuited to present a patient journey.

Mendel’s Hypercube represents a paradigm shift, mapping the intricate, multi-dimensional relationships inherent in clinical information.
Photo of man talking with two other people at a desk with a laptop on the desk.
Man writing on a whiteboard window

It can scale with medicine’s first Generative Ontology.

Existing approaches rely on expert curation of data making them difficult to scale. It also relies on current ontologies which do not fit reasoning. Each ontology was tailored for some to serve a unique purpose, encodings overlap unsystematically and are often in conflict.

Mendel’s Object-Oriented Ontology is medicine's first generative ontology—a concept that combines principles of generative grammar and ontology to create systems that can:
  • Adapt to new information or contexts, offering a scalable way to model medicine.
  • Capture both the structure and the dynamics of the clinical domain.
  • Interoperate and map to standard ontologies.

And works at the speed of thought

Traversing a hypergraph introduces unique challenges compared to more traditional graph structures, largely due to the complexity and flexibility of hypergraphs. Mendel's developed innovative algorithms and data structures designed specifically for hypergraph analysis. It can answer questions in real-time no matter the number of nodes and relations involved.
Photo of people working at the Mendel office in San Jose looking and computers.
Photo of Dung Thai, Senior AI Research Scientist at Mendel.ai.

Our LLM approach: Survival of the fittest.

Our approach to leveraging Large Language Model prioritizes portability. Models are commodities one should always be able to leverage the best while prioritizing precise data selection, expert domain-specific annotation teams, and customized annotation tasks. Our team of AI scientists and clinical experts designing the right annotation task, curating training data and leverage real medical records
What Our Customers Say

“While I have decades of experience analyzing large clinical data sets, the ability to simply ‘ask’ Hypercube to find data of interest and to analyze all structured and unstructured data is a quantum leap.”

Senior Director of Bioinformatics at A top 50 Life Science company

“The results here suggest that domain specific models can outperform generic LLMs at scale”

Leading Breast Cancer Oncologist, Vice Provost, Top 5 University Medical Center

Case Studies and Publications

Transforming Unstructured Pathology Data
Read
→
Mendel Unveils Groundbreaking Neuro-Symbolic AI System Outperforming GPT-4 for Automatic Cohort Retreival in New Study
Read
→
Improving Clinical Trial Participant Prescreening With Artificial Intelligence (AI): A Comparison of the Results of AI Assisted vs Standard Methods in 3 Oncology Trials
Read
→
View Success Stories

Get started with the
Mendel platform today

Try the Playground
Contact Sales