37

GeMorph: Generating 3D Facial Models from Genomic Data

GeMorph: Generating 3D Facial Models from Genomic Data

Law enforcement agencies in Pakistan face substantial backlogs, processing times of 4–6 months in murder cases and conviction rates of only 0.2% in rape cases, due to resource constraints and database limitations. Meanwhile, generating accurate phenotypic predictions from high-dimensional genomic data presents challenges in model training, data sparsity for non-European ancestries, and privacy/security of sensitive genetic information.

GeMorph addresses a critical gap in forensic investigations by transforming raw genomic data into fully textured 3D facial models. Traditional DNA profiling methods rely on matching samples against existing databases, but when no match is found, investigations often stall. GeMorph overcomes this limitation by leveraging SNP analysis, advanced machine-learning pipelines (including classification models for eye, hair, and skin pigmentation, and diffusion-based facial generation), and texture-mapping techniques to produce 3D facial composites directly from DNA samples.

GeMorph’s workflow begins with DNA data preprocessing, including quality control and strand correction. Separate classification pipelines then process this data to predict traits like eye color, hair color, skin tone, gender, and ancestry, providing categorical predictions and confidence scores. A conditional autoencoder model then uses these outputs, along with the DNA data, to refine an average-face template and generate individualized 3D facial meshes, which are texture-mapped to reflect the predicted pigmentation traits. The system is integrated into a secure offline desktop application for on-premises use, with a liaison website for case intake and tracking. Unlike commercial ‘black box’ services (Parabon Snapshot), GeMorph is built for transparency and extensibility, with modular, auditable pipelines. It delivers a unified system for phenotype prediction and 3D face generation, producing a comprehensive report that compiles all predictions in a clear, interpretable format.

Faculty

Students

  • Syed Ahsan Ullah Tanweer
  • Navaira Rehman
  • Syeda Fatima Roshan Haneef (ASAB)

Comments are closed.