Yu S. Huang

Multimodal epigenetic characterization of cell-free DNA (cfDNA) could improve the performance of blood-based early cancer detection. However, integrative profiling of cfDNA methylome and fragmentome has been technologically challenging. Here, we adapt an enzyme-mediated methylation sequencing method for comprehensive analysis of genome-wide cfDNA methylation, fragmentation, and copy number alteration (CNA) characteristics for enhanced cancer detection. We apply this method to plasma samples of 497 healthy controls and 780 patients of seven cancer types and develop an ensemble classifier by incorporating methylation, fragmentation, and CNA features. In the test cohort, our approach achieves an area under the curve value of 0.966 for overall cancer detection. Detection sensitivity for early-stage patients achieves 73% at 99% specificity. Finally, we demonstrate the feasibility to accurately localize the origin of cancer signals with combined methylation and fragmentation profiling of tissue-specific accessible chromatin regions. Overall, this proof-of-concept study provides a technical platform to utilize multimodal cfDNA features for improved cancer detection.

eGADA is an enhanced version of GADA, which is a fast segmentation algorithm utilizing the Sparse Bayesian Learning (or Relevance Vector Machine) technique from Tipping 2001. It can be applied to array intensity data, NGS sequencing coverage data, or any sequential data that displays characteristics of stepwise functions. Improvements by eGADA over GADA include: a) a customized Red-Black tree to significantly expedite the final backward elimination step of GADA; b) code in C++, which is safer and better structured than C; c) use Boost libraries extensively to provide user-friendly help and commandline argument processing; d) user-friendly input and output formats; e) export a dynamic library eGADA.so (packaged via Boost.Python) that offers API to Python; f) some bug fixes/optimization. The code is published at https://github.com/polyactis/eGADA.

Deep learning-based virtual screening methods have been shown to significantly improve the accuracy of traditional docking-based virtual screening methods. In this paper, we developed Deffini, a structure-based virtual screening neural network model. During training, Deffini learns protein-ligand docking poses to distinguish actives and decoys and then to predict whether a new ligand will bind to the protein target. Deffini outperformed Smina with an average AUC ROC of 0.92 and AUC PRC of 0.44 in 3-fold cross-validation on the benchmark dataset DUD-E. However, when tested on the maximum unbiased validation (MUV) dataset, Deffini achieved poor results with an average AUC ROC of 0.517. We used the family-specific training approach to train the model to improve the model performance and concluded that family-specific models performed better than the pan-family models. To explore the limits of the predictive power of the family-specific models, we constructed Kernie, a new protein kinase dataset consisting of 358 kinases. Deffini trained with the Kernie dataset outperformed all recent benchmarks on the MUV kinases, with an average AUC ROC of 0.745, which highlights the importance of quality datasets in improving the performance of deep neural network models and the importance of using family-specific models.

BACKGROUND: Copy number alterations (CNAs), due to their large impact on the genome, have been an important contributing factor to oncogenesis and metastasis. Detecting genomic alterations from the shallow-sequencing data of a low-purity tumor sample remains a challenging task. RESULTS: We introduce Accucopy, a method to infer total copy numbers (TCNs) and allele-specific copy numbers (ASCNs) from challenging low-purity and low-coverage tumor samples. Accucopy adopts many robust statistical techniques such as kernel smoothing of coverage differentiation information to discern signals from noise and combines ideas from time-series analysis and the signal-processing field to derive a range of estimates for the period in a histogram of coverage differentiation information. Statistical learning models such as the tiered Gaussian mixture model, the expectation–maximization algorithm, and sparse Bayesian learning were customized and built into the model. Accucopy is implemented in C++ /Rust, packaged in a docker image, and supports non-human samples, more at http://www.yfish.org/software/. CONCLUSIONS: We describe Accucopy, a method that can predict both TCNs and ASCNs from low-coverage low-purity tumor sequencing data. Through comparative analyses in both simulated and real-sequencing samples, we demonstrate that Accucopy is more accurate than Sclust, ABSOLUTE, and Sequenza.