Aktiver/CoreTX-Agent

In-silico Experiments with Molecular Interactions for Personalized Treatments

Aktiver CoreTx Agent enables clinician to run billions of in-silico experiments for precise therapeutic exploration grounded in real-time reasoning, not outdated parameters. By linking patient phenotypes to molecular targets and drug mechanisms, Aktiver CoreTx simulates personalized interventions across pharmacokinetic, genomic, and clinical dimensions. Built on a foundation of 190+ verified tools—including openFDA, Open Targets, and the Human Phenotype Ontology—it dynamically retrieves and sequences biomedical knowledge to evaluate efficacy, contraindications, and synergistic effects. Trained on over 150K clinical reasoning traces and 250K function calls, it applies multi-step logic to support treatment design and drug repurposing & discovery in a multitude of complex diseases. The result is an agentic system that merges LLM-driven planning with structured RDF knowledge graphs to deliver transparent, personalized therapeutics—before a single dose is administered.

Aktiver CoreTx Agent: Real-Time Therapeutic Reasoning with Verified Tools

Precision care demands adaptive, multimodal agents that personalize therapy decisions. Aktiver CoreTx Agent is an AI system built for high-stakes clinical reasoning. It combines structured function calls, real-time biomedical knowledge retrieval, and multi-step logic to evaluate drug mechanisms, contraindications, and patient-specific treatment plans.

CoreTx integrates 190 tools across pharmacology, genetics, and clinical guidelines, retrieving and synthesizing current biomedical data to guide decision-making. These tools come from verified sources, including openFDA [1], Open Targets [2], and the Human Phenotype Ontology [3].

It evaluates drugs on pharmacokinetic, molecular, and clinical dimensions, checking comorbidities, age, genetics, and drug interactions in context. CoreTx applies iterative reasoning: selecting tools based on the task, refining answers step-by-step, and ensuring compliance with guidelines and up-to-date evidence.

Architecture and Toolchain

Clincial Lab Tools: a curated toolkit of over 190+ biomedical utilities, that chooses tools dynamically.

A fine-tuned reasoning LLM: trained for multi-step planning and tool use.

New tools are constructed automatically using the multi-agent system that parses API documentation. The training comprises over 300k examples, including over 150k clinician reasoning steps and over 250k function calls sourced from FDA records.

These tools integrate trusted sources, including openFDA [1], Open Targets [2], and the Human Phenotype Ontology [3].

1. Kass-Hout et al. Journal of the American Medical Informatics Association, 2016. PDF

2. Ochoa et al. Nucleic Acids Research, 2023. PDF

3. Castellanos et al. Nucleic Acids Research, 2024. PDF

Performance Across 3 Clinical Benchmarks

Aktiver CoreTx outperforms GPT-4o and other LLMs across five new evaluations:

Rx Drugs, Treatment Modalities, Clinical Descriptors

• 3K+ drug tasks and 400+ treatment cases
• 90% open-ended drug reasoning accuracy
• Beats GPT-4o by 25% and Llama3.1-70B by 40%

Accuracy variance under 0.01 across brand, generic, and descriptive drug inputs

In real-world tests, Aktiver CoreTx excels in:

Context-based drug identity inference

• Personalizing treatment decisions
• Navigating brand/generic naming inconsistencies
• Handling novel FDA approvals post-model training

Real-Time Reasoning + Verified Knowledge Tracing on RDF Graphs

Unlike static LLMs, CoreTx is dynamic. It doesn't rely on outdated weights—it calls tools in real time. When a patient-specific query is posed, CoreTx:

• Uses validated decision making to select the best tools
• Generates reasoning steps with corresponding in-silico experiments
• Synthesizes responses with transparent justification

Adaptive, Modular, and Grounded

The system evolves as new APIs and datasets emerge. Whether finding adverse reactions for novel drug combos or matching diseases to protein targets, CoreTx:

• Forms plans
• Executes tool chains
• Adjusts based on feedback
• Outputs final answers and full reasoning traces

It supports clinical professionals, regulators, and researchers in navigating ever-changing biomedical data landscapes with accountability and transparency.