PHARMore

• 

  PDF annotation functionality

• 

  Drug Discovery

Inspiration
We were driven by the need to reduce the lengthy processes in drug annotation and testing. By streamlining these tasks, researchers can focus more on critical analysis and innovation in drug development.

What it does
PHARMore is an all-in-one application for drug discovery and medical research. Its functionality includes:

1. When given a pdf or a url link of a research paper or report, it can extract detailed drug mentions and metadata and provides recommendations for similar drugs
2. A chatbot which can answer medical questions from the user about a drug or a disease
3. When given the PubChem ID of a drug, provide alternative drugs with a similar structure for drug repurposing These functionalities assists researchers in quickly identifying potential drug candidates and alternative therapies.

How we built it
We utilized public databases to identify drug mentions and relevant metadata, then built machine learning models that analyze these details to return recommendations on similar drugs. The process involved integrating natural language processing techniques with robust ML algorithms to ensure accuracy and relevance. In our case, we utilized the Google Gemini API to create a medically informed LLM. We also used Streamlit to deploy our web application, and used AWS to host the app on a .tech domain.

Challenges we ran into
Finding viable datasets proved to be a significant hurdle. We had to address issues like inconsistent data formats, incomplete records, and the need for effective normalization of the extracted metadata to ensure the models could learn efficiently. Biomedical data is also very diverse and each dataset has its own niche, which made it difficult to create a one-size fits all solution.

Accomplishments that we're proud of
We successfully built a working demo that not only extracts key drug information from complex medical texts but also provides reliable recommendations for similar drugs. This demo validates the feasibility of integrating literature mining with ML in real-world pharmaceutical research.

What we learned
Throughout this project, we picked up a range of practical skills that made a real difference. We became much more comfortable with the nitty-gritty of cleaning and normalizing large datasets—a critical step when dealing with complex medical reports. Working closely with natural language processing and machine learning, we learned how to extract valuable insights from messy real-world data. Plus, the hands-on experience with model tuning and troubleshooting has really prepared us for future challenges in data-driven drug discovery.

What's next for PHARMore
We plan to expand our focus beyond drug similarity recommendations. Future developments include incorporating additional pharmaceutical applications, such as adverse effect prediction, dosage optimization, and personalized medicine insights, further empowering medical research and drug discovery.

Built With

• amazon-web-services
• gemini
• llm
• machine-learning
• python
• streamlit
• torch