DeepOM: Single-molecule optical genome mapping via deep learning

Abstract

Efficient tapping into genomic information from a single microscopic image of an intact DNA molecule fragment is an outstanding challenge and its solution will open new frontiers in molecular diagnostics. Here, a new computational method for optical genome mapping utilizing Deep Learning is presented, termed DeepOM. Utilization of a Convolutional Neural Network (CNN), trained on simulated images of labeled DNA molecules, improves the success rate in alignment of DNA images to genomic references. The method is evaluated on acquired images of human DNA molecules stretched in nano-channels. The accuracy of the method is benchmarked against state-of-the-art commercial software Bionano Solve. The results show a significant advantage in alignment success rate for molecules shorter than 50 kb. DeepOM improves yield, sensitivity and throughput of Optical DNA Mapping experiments in applications of human genomics and microbiology.

Design of optimal labeling patterns for optical genome mapping via information theory

“…This data consisted of images acquired as described in Section 2.4, where the DNA fragments were labeled by the DLE-1 enzyme and imaged using the Saphyr system (Bionano Genomics). The DNA fragments, all longer than 400kb, were aligned to the human genome, using the DeepOM algorithm (Nogin et al, 2023). Overall, 445 DNA fragments were used consisting of a total of 180 megabase (Mb).…”

“…The experimental validation of the theory in Equation ( 1) was done by comparing it to results from the DeepOM work (Nogin et al, 2023), which evaluated the error rates of the DeepOM OGM algorithm in mapping DNA fragments to the genome, using experimental human DNA fragments labeled with the pattern CTTAAG (Figure 1). The algorithm is based on a Convolutional Neural Network (CNN) for localizing labels in a DNA fragment image, and a Dynamic Programming (DP) algorithm for aligning the DNA fragment to the genome.…”

“…These images were captured using the Saphyr instrument and Saphyr chips (G1.2) (Bionano Genomics). The protocol for extraction and labeling of DNA is described in Nogin et al (Nogin et al, 2023). U2OS human cell line was cultured in Dulbecco's Modified Eagle medium with 10% fetal bovine serum (Gibco, Amarillo, TX), 2 mM l-glutamine, and 1% penicillin-streptomycin (10,000 U/mL; Gibco).…”

“…We validate the theory through simulations as described in Section 2.2. For the human genome, we also compare to the experimental data from DeepOM (Nogin et al, 2023).…”

This microtubule does not exist: Super-resolution microscopy image generation by a diffusion model

“…Deep learning algorithms have been extensively used in the past decade to solve various microscopy challenges [1][2][3][4][5][6][7] . These algorithms outperform traditional computer vision methods in terms of reconstruction quality, analysis time, and classification, among many others.…”