EN KR
Linkedin Facebook

Discovery & Development

Hilbert CNN-PGM : Biomarker discovery and Patient Stratification

This technology transforms genomic information into image and utilizes artificial intelligence for learning to analyze the complex relationships within the genome in a non-linear function, which are difficult to prove through human computation.

SCROLL DOWN

Solution

How does the Hilbert-CNN approach enhance the analysis and prediction of complex diseases using genomic data?

Hilbert-CNN uses the Hilbert curve to turn genomic data into images, capturing nonlinear relationships. And then analyzes these images to learn and infer patterns, improving disease classification.

How does Hilbert-CNN address reproducibility issues when applying biomarkers across different cohorts?

Hilbert-CNN achieves over 90% reproducibility through blind set validation, outperforming traditional methods which typically show around 50-60% performance on blind datasets.

Workflow

A
  • ADIscan manages data quality before Hilbert Curve transforms SNP data into images. Color Mapping assigns colors to genotypes
B
  • CNN Model is trained on these images for clarification, with Cross-Validation for performance and integrated gradients for prediction.
C
  • The process ends with summary model generation, using data splits, probabilities, AUC scores, and normalization for performance assessment,

Application

  • AI discrimination models utilizing genomic data can go beyond simple disease presence detection.
  • Customized models can discern complex diseases, like cancer, based on specific criteria.
  • They're applicable in medical scenarios, such as cancer type differentiation, metastasis detection, and patient prognosis prediction, with diverse classification models developed for various conditions.
References:
  • Cho et al. “Development of the variant calling algorithm, ADIScan, and its use to estimate discordant sequences between monozygotic twins” Nucleic Acids Research. 2018 Sep 6;46(15):e92.
  • Hilbert-cnn: ai-driven convolutional neural networks with conversion data of genome for biomarker discovery. USA, 16-198201. 2018
  • SYSTEM AND METHOD FOR ANALYZING GENOTYPE USING GENETIC VARIATION INFORMATION ON INDIVIDUAL'S GENO

send us your inquiry

SUBMIT
SUBMIT