NexJen Bio

• 

  Team Logo

Inspiration

The pharmaceutical industry spends on average \$2.6 , \text{B} per new drug, yet fewer than 10% of compounds succeed in clinical trials. This inefficiency stems from unreliable forecasting, fragmented datasets, and lack of accessible, explainable intelligence. We were inspired to tackle this gap by building a platform that acts like an autonomous AI analyst team, capable of unifying biomedical, clinical, and financial signals into actionable insights.

What it does

NexJen Bio is an agentic AI-powered platform that predicts clinical trial outcomes, evaluates business development opportunities, and forecasts IPO readiness. It is designed for pharma, biotech firms, CROs, and investors.

Key modules include:

• Clinical Trial Insight Hub – Predicts trial success rates, enrollment risks, and trial quality.
• BD & IPO Scoring Platform – Assesses pipeline strength, financial health, and IPO timing.
• Paper & Signal Feed – A mobile feed delivering personalized research and trial intelligence.
• Stock Investment Suite – Provides risk-adjusted insights for biotech market investments.

Together, these modules transform raw data into explainable, real-time recommendations, helping stakeholders cut waste, reduce risk, and accelerate life-saving therapies.

How we built it

We designed NexJen Bio as a multi-agent system orchestrated with IBM watsonx:

• Retriever Agent: Continuously ingests data from ClinicalTrials.gov, PubMed, SEC filings, and market feeds.
• Analyst Agent: Runs predictive ML models: GBDT for \(\text{PoS}\), GNN for target–disease graphs, and stacked LSTM–XGBoost for IPO forecasting.
• Synthesizer Agent: Uses retrieval-augmented generation (RAG) to generate concise summaries of mechanisms, safety, and KOL sentiment.
• Advisor Agent: Adapts insights to user personas (pharma BD, investors, CROs).

Mathematically, our success probability model is expressed as:

$$ \text{PoS} = f_{\text{GBDT}}(X_{\text{clinical}}) + f_{\text{GNN}}(X_{\text{target-disease}}) + f_{\text{LSTM-XGB}}(X_{\text{market}}) $$

Challenges we ran into

• Unifying heterogeneous biomedical, financial, and regulatory datasets into one scalable pipeline.
• Balancing prediction accuracy with explainability in regulated domains.
• Managing real-time data ingestion without excessive cloud costs.
• Designing interfaces intuitive enough for both scientists and investors.

Accomplishments that we're proud of

• Built a functioning multi-agent AI prototype integrated with IBM watsonx for explainability and governance.
• Demonstrated trial success predictions with accuracy competitive to leading academic benchmarks.
• Created a mobile feed app that combines ArXiv-style literature with ClinicalTrials intelligence.
• Piloted the system with real-world trial datasets and achieved positive feedback from early biotech partners.

What we learned

• Agentic AI architectures are highly effective in biomedical decision-making when paired with governance frameworks like watsonx.
• Building trust in AI for healthcare requires more than accuracy—it requires transparency, auditability, and domain alignment.
• A key differentiator is not just predictions, but how insights are delivered—in user-friendly, actionable formats.

What's next for NexJen Bio

• Expand dataset coverage to include real-world evidence (EHRs, claims data).
• Enterprise pilots with pharma and CRO partners to validate business value.
• Scale commercialization by onboarding VC firms and biotech investment funds.
• Incorporate reinforcement learning agents that dynamically adjust trial/market forecasts as new data arrives.
• Continue leveraging IBM watsonx.ai, watsonx.data, and watsonx.governance to ensure accuracy, scalability, and compliance.

Built With

• amazon-web-services
• bilibili
• biorxiv
• docker
• financial-market-feeds-frontend:-react
• hugging-face-transformers-cloud-&-infrastructure:-ibm-watsonx.ai
• ibm-watsonx.data
• ibm-watsonx.governance
• jupyter
• kubernetes-databases-&-apis:-clinicaltrials.gov-api
• languages-&-frameworks:-python
• networkx
• pubmed-api
• pytorch
• restful-apis
• scikit-learn
• sec-edgar
• streamlit-(demo-prototyping)
• tailwindcss-other-tools:-github
• xgboost