Deep learning integrates histopathology and proteogenomics at a pan-cancer level.

We introduce a pioneering approach that integrates pathology imaging with transcriptomics and proteomics to identify predictive histology features associated with critical clinical outcomes in cancer. We utilize 2,755 H&E-stained histopathological slides from 657 patients across 6 cancer types from CPTAC. Our models effectively recapitulate distinctions readily made by human pathologists: tumor vs. normal (AUROC = 0.995) and tissue-of-origin (AUROC = 0.979). We further investigate predictive power on tasks not normally performed from H&E alone, including TP53 prediction and pathologic stage. Importantly, we describe predictive morphologies not previously utilized in a clinical setting. The incorporation of transcriptomics and proteomics identifies pathway-level signatures and cellular processes driving predictive histology features. Model generalizability and interpretability is confirmed using TCGA. We propose a classification system for these tasks, and suggest potential clinical applications for this integrated human and machine learning approach. A publicly available web-based platform implements these models.

Cell-type-specific prediction of 3D chromatin organization enables high-throughput in silico genetic screening.

Investigating how chromatin organization determines cell-type-specific gene expression remains challenging. Experimental methods for measuring three-dimensional chromatin organization, such as Hi-C, are costly and have technical limitations, restricting their broad application particularly in high-throughput genetic perturbations. We present C.Origami, a multimodal deep neural network that performs de novo prediction of cell-type-specific chromatin organization using DNA sequence and two cell-type-specific genomic features-CTCF binding and chromatin accessibility. C.Origami enables in silico experiments to examine the impact of genetic changes on chromatin interactions. We further developed an in silico genetic screening approach to assess how individual DNA elements may contribute to chromatin organization and to identify putative cell-type-specific trans-acting regulators that collectively determine chromatin architecture. Applying this approach to leukemia cells and normal T cells, we demonstrate that cell-type-specific in silico genetic screening, enabled by C.Origami, can be used to systematically discover novel chromatin regulation circuits in both normal and disease-related biological systems.

Modeling the Heating Dynamics of a Semiconductor Bridge Initiator with Deep Neural Network.

A semiconductor bridge (SCB) is an ignition device that provides a safe and efficient method widely used in civilian and military fields. The heating process of an SCB under electrical stimulation has a wide range of applications owing to its unique energy release process. However, the temperature variation of an SCB is challenging to obtain, both experimentally because of the rapid reaction on a microscale and with simulation due to its high demand in nonlinear calculations. In this study, we propose deep learning (DL) approach to study the electrothermal-coupled multi-physical heating process of the SCB initiator. We generated training data with multi-physics simulation (MPS), producing surface temperature distributions of SCBs under different voltages. The model was then trained with partial data in this database and evaluated on a separate test set. A generative adversarial network (GAN) with a customized loss function was used for modeling point-wise temperature dynamics. In the test set, our proposed method can predict the temperature distribution of an SCB under different voltages with high accuracy of over 0.9 during the heating process. We reduced the computation time by several orders of magnitude by replacing MPS with a deep neural network.

Prediction of Total Knee Replacement and Diagnosis of Osteoarthritis by Using Deep Learning on Knee Radiographs: Data from the Osteoarthritis Initiative.

Background The methods for assessing knee osteoarthritis (OA) do not provide enough comprehensive information to make robust and accurate outcome predictions. Purpose To develop a deep learning (DL) prediction model for risk of OA progression by using knee radiographs in patients who underwent total knee replacement (TKR) and matched control patients who did not undergo TKR. Materials and Methods In this retrospective analysis that used data from the OA Initiative, a DL model on knee radiographs was developed to predict both the likelihood of a patient undergoing TKR within 9 years and Kellgren-Lawrence (KL) grade. Study participants included a case-control matched subcohort between 45 and 79 years. Patients were matched to control patients according to age, sex, ethnicity, and body mass index. The proposed model used a transfer learning approach based on the ResNet34 architecture with sevenfold nested cross-validation. Receiver operating characteristic curve analysis and conditional logistic regression assessed model performance for predicting probability and risk of TKR compared with clinical observations and two binary outcome prediction models on the basis of radiographic readings: KL grade and OA Research Society International (OARSI) grade. Results Evaluated were 728 participants including 324 patients (mean age, 64 years ± 8 [standard deviation]; 222 women) and 324 control patients (mean age, 64 years ± 8; 222 women). The prediction model based on DL achieved an area under the receiver operating characteristic curve (AUC) of 0.87 (95% confidence interval [CI]: 0.85, 0.90), outperforming a baseline prediction model by using KL grade with an AUC of 0.74 (95% CI: 0.71, 0.77; P < .001). The risk for TKR increased with probability that a person will undergo TKR from the DL model (odds ratio [OR], 7.7; 95% CI: 2.3, 25; P < .001), KL grade (OR, 1.92; 95% CI: 1.17, 3.13; P = .009), and OARSI grade (OR, 1.20; 95% CI: 0.41, 3.50; P = .73). Conclusion The proposed deep learning model better predicted risk of total knee replacement in osteoarthritis than did binary outcome models by using standard grading systems. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Richardson in this issue.

Comprehensive codon usage analysis of porcine deltacoronavirus.

Porcine deltacoronavirus (PDCoV) is a newly identified coronavirus of pigs that was first reported in Hong Kong in 2012. Since then, many PDCoV isolates have been identified worldwide. In this study, we analyzed the codon usage pattern of the S gene using complete coding sequences and complete PDCoV genomes to gain a deeper understanding of their genetic relationships and evolutionary history. We found that during evolution three groups evolved with a relatively low codon usage bias (effective number of codons (ENC) of 52). The factors driving bias were complex. However, the primary element influencing the codon bias of PDCoVs was natural selection. Our results revealed that different natural environments may have a significant impact on the genetic characteristics of the strains. In the future, more epidemiological surveys are required to examine the factors that resulted in the emergence and outbreak of this virus.

Analysis of codon usage bias in the VP1 gene of the human norovirus GII.2 genotype.

To investigate the codon usage patterns of all available VP1 gene sequences of the GII.2 genotype, to determine the factors that affect these patterns, and to provide comprehensive details of the characteristics and evolution of the gene. Complete 519 sequences of VP1 gene of the HuNoV GII.2 genotype with known sampling dates and geographic locations from 1971 - 2017 were retrieved from the GenBank nucleotide database of the National Center for Biotechnology Information (NCBI) and analyzed. The percentage composition of T, C, A, and G nucleotides were 24.80 ± 0.30, 26.61 ± 0.31, 25.84 ± 0.13, and 22.75 ± 0.17 %, respectively, with C and A relatively more abundant than T and G, and C the most abundant (p < 0.0001). The values of T3s (34.10 ± 0.90 %) and C3s (33.54 ± 0.90 %) were significantly higher than those of A3s (29.98 ± 0.43 %) and G3s (24.13 ± 0.51 %) (p < 0.0001). While T3s was highest among the four nucleotides, G3s was the lowest. Among the 18 most frequently employed synonymous codons, six optional codons ended with T, five ended with C, five ended with A and two ended with G. Codons ending with T were the most frequently used. The ENC ranged from 51.90 to 54.25 (mean = 52.38 ± 0.43) among the 519 VP1 gene sequences. There were significant correlations between ENC and C % and G % (p < 0.01). Codons containing CpG (1 and 2 or 2 and 3 codon positions) showed the lowest frequencies, while 30, 29, and 2 codons were above, below and on the mean line, respectively. The first four principal components accounted for 69.11 % of the total variation, with the first, second, third, and fourth principal axes contributing 37.90, 14.83, 9.61, and 6.77 %, respectively. The strains were not clustered by country of isolation or year of sampling. Gravy were significantly correlated with T3s, C3s, G3s, GC3s, and ENC (p < 0.01). Mutation pressure and natural selection contributed to the codon usage bias of the VP1 gene of the HuNoV GII.2 genotype. There was a correlation between GC12s and GC3s (R2 = 0.032; p < 0.0001). The relative neutrality was 3.20 %, while natural selection was 96.80 %. The VP1 gene exhibits low codon usage bias which is affected primarily by natural selection, followed by mutation pressure and translational selection.

A Novel Trypsin Inhibitor-Like Cysteine-Rich Peptide from the Frog Lepidobatrachus laevis Containing Proteinase-Inhibiting Activity.

Various bio-active substances in amphibian skins play important roles in survival of the amphibians. Many protease inhibitor peptides have been identified from amphibian skins, which are supposed to negatively modulate the activity of proteases to avoid premature degradation or release of skin peptides, or to inhibit extracellular proteases produced by invading bacteria. However, there is no information on the proteinase inhibitors from the frog Lepidobatrachus laevis which is unique in South America. In this work, a cDNA encoding a novel trypsin inhibitor-like (TIL) cysteine-rich peptide was identified from the skin cDNA library of L. laevis. The 240-bp coding region encodes an 80-amino acid residue precursor protein containing 10 half-cysteines. By sequence comparison and signal peptide prediction, the precursor was predicted to release a 55-amino acid mature peptide with amino acid sequence, IRCPKDKIYKFCGSPCPPSCKDLTPNCIAVCKKGCFCRDGTVDNNHGKCVKKENC. The mature peptide was named LL-TIL. LL-TIL shares significant domain similarity with the peptides from the TIL supper family. Antimicrobial and trypsin-inhibitory abilities of recombinant LL-TIL were tested. Recombinant LL-TIL showed no antimicrobial activity, while it had trypsin-inhibiting activity with a Ki of 16.5178 µM. These results suggested there was TIL peptide with proteinase-inhibiting activity in the skin of frog L. laevis. To the best of our knowledge, this is the first report of TIL peptide from frog skin.