Inspiration
Antimicrobial resistance (AMR) is an urgent global health issue that threatens the effectiveness of life-saving treatments. Traditional Multiple Sequence Alignment (MSA) techniques are computationally expensive and often struggle with accuracy. Inspired by the Borderlands Science project, which uses human problem-solving strategies in bioinformatics, we set out to combine AI with human intuition to optimize sequence alignment for AMR detection.
What We Built
We developed AlignAI, an AI-powered tool designed to predict optimal sequence alignments based on human-generated data. Our solution involves:
Data Processing: Extracting human gameplay data from Borderlands Science. Machine Learning Model: Training a Random Forest classifier to predict optimal alignment strategies. Optimization: Automating the best sequence moves to maximize the alignment score. Evaluation: Testing AI-generated solutions against human-curated alignments. Challenges We Ran Into
Computational Complexity: MSA is an NP-hard problem, making optimization non-trivial. Feature Engineering: Converting human move sequences into a structured dataset required creative preprocessing. Generalization Issues: AI sometimes overfit to training data, making it harder to perform well on unseen sequences. Scoring Function Optimization: Aligning AI-generated results with biologically relevant alignments was a significant challenge. Accomplishments That We're Proud Of
Successfully trained an AI model that mimics human intuition for sequence alignment. Optimized alignment predictions, reducing computational cost while improving accuracy. Demonstrated AI-human collaboration in a bioinformatics challenge. Presented a novel approach to tackling AMR detection through AI-driven sequence alignment. What We Learned
The power of crowdsourcing: Human problem-solving strategies can complement AI models in complex tasks. AI’s role in bioinformatics: Machine learning can enhance sequence alignment, but fine-tuning is essential. Balancing performance & interpretability: Some AI models work well but lack biological relevance—context matters. Iterative model improvements: Small adjustments in feature selection and hyperparameters significantly impact results.
What’s Next for AlignAI
AlignAI is just the beginning! Future improvements include:
Deep Learning Integration: Experimenting with transformers and reinforcement learning for better sequence prediction. Expanded Training Data: Incorporating more diverse datasets to improve generalization. Interactive AI-assisted Alignment: Developing a user-friendly interface where researchers can interact with AI-suggested alignments. Scaling Up: Deploying AlignAI to assist bioinformatics labs in real-world AMR research.
Log in or sign up for Devpost to join the conversation.