OUR PROJECTS
De Novo Drug Design
Diffusion models are a cutting-edge generative approach in structure-based drug design, capable of creating novel molecular structures by iteratively refining random noise into meaningful chemical configurations. These models excel at exploring vast chemical spaces while leveraging structural data of target proteins. A critical consideration is ensuring physical plausibility, as generated molecules must not only fit within the target binding pocket but also adhere to principles of chemistry, such as realistic bond lengths, angles, and steric interactions. Ensuring physical plausibility is critical for seamless progression from in silico design to in vitro testing and experimental validation.
Music Genre Classifier
Building an accurate music genre classifier involves leveraging audio features that capture the unique characteristics of different genres. Key steps include extracting features such as tempo, pitch, timbre, and rhythm using tools like Mel-Frequency Cepstral Coefficients or spectrograms. After comparing various deep learning methods and analyzing feature importance using random forests, XGBoost emerged as the optimal choice due to its balance of accuracy, interpretability, and computational efficiency. Ensuring a diverse and well-labeled training dataset is crucial for capturing genre variability and avoiding bias. Additionally, careful model evaluation using metrics like F1 score ensures the classifier performs well across all genres, including underrepresented ones.
Activity GPX Visualizer
A Python function was developed to process GPX files and generate interactive maps using Plotly. This tool dynamically visualizes GPS data, including paths, waypoints, and elevation profiles, in a user-friendly format. It has practical applications, such as enhancing the experience for participants in biking or running events by allowing them to explore the event route interactively. Additionally, it can be used by outdoor enthusiasts to plan trips, share routes, or analyze performance data from past activities.
FFAW R Library
A custom R library was developed to support FFAW-Unifor, focusing on streamlining data analysis and reporting workflows for the organization. The library includes tailored functions for processing fisheries data, generating summary statistics, and creating visually appealing, publication-ready plots. By automating repetitive tasks and ensuring consistency in analyses, the library enhances productivity and enables more efficient communication of insights to stakeholders. This tool is particularly valuable for addressing the unique challenges of fisheries management and policy development.