DDBJ Data Analysis Challenge: a machine learning competition to predict Arabidopsis chromatin feature annotations...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Automating annotation analysis of large-scale next-generation sequencing data using machine learning (ML) is of high interest.
• Experimental life scientists often face challenges in finding ML collaborators.
• Crowdsourcing offers a potential solution to bridge this gap.

Framework:

• Investigated crowdsourced modeling for life science tasks via a machine learning competition: the DNA Data Bank of Japan (DDBJ) Data Analysis Challenge.
• Participants predicted chromatin feature annotations from DNA sequences using competing ML models.
• The challenge involved 38 participants and 360 model submissions.

Implementation:

• The top-performing model achieved an Area Under the Curve (AUC) score of 0.95.
• Overall model performance improved by an AUC of 0.30 during the competition.
• Top models incorporated external data, such as genomic location and gene annotations, demonstrating the value of domain knowledge.

Implications:

• Crowdsourced ML competitions can develop highly accurate models for experimental scientists lacking data science expertise.
• Incorporating domain knowledge into ML models led to significant performance improvements (5%-9% AUC).
• This approach democratizes access to advanced computational tools in life sciences research.