In the era of precision medicine, cancer therapy can be tailored to an individual patient based on the genomic profile of a tumour. Despite the ever-increasing abundance of cancer genomic data, linking mutation profiles to drug efficacy remains a challenge. Herein, we report Cancer Drug Response profile scan (CDRscan) a novel deep learning model that predicts anticancer drug responsiveness based on a large-scale drug screening assay data encompassing genomic profiles of 787 human cancer cell lines and structural profiles of 244 drugs. CDRscan employs a two-step convolution architecture, where the genomic mutational fingerprints of cell lines and the molecular fingerprints of drugs are processed individually, then merged by ‘virtual docking’, an in silico modelling of drug treatment. Analysis of the goodness-of-fit between observed and predicted drug response revealed a high prediction accuracy of CDRscan (R² > 0.84; AUROC > 0.98). We applied CDRscan to 1,487 approved drugs and identified 14 oncology and 23 non-oncology drugs having new potential cancer indications. This, to our knowledge, is the first-time application of a deep learning model in predicting the feasibility of drug repurposing. By further clinical validation, CDRscan is expected to allow selection of the most effective anticancer drugs for the genomic profile of the individual patient.

Several recent studies showed that next-generation sequencing (NGS)-based human leukocyte antigen (HLA) typing is a feasible and promising technique for variant calling of highly polymorphic regions. To date, however, no method with sufficient read depth has completely solved the allele phasing issue. In this study, we developed a new method (HLAscan) for HLA genotyping using NGS data. HLAscan performs alignment of reads to HLA sequences from the international ImMunoGeneTics project/human leukocyte antigen (IMGT/HLA) database. The distribution of aligned reads was used to calculate a score function to determine correctly phased alleles by progressively removing false-positive alleles. Comparative HLA typing tests using public datasets from the 1000 Genomes Project and the International HapMap Project demonstrated that HLAscan could perform HLA typing more accurately than previously reported NGS-based methods such as HLAreporter and PHLAT. In addition, the results of HLA-A, −B, and -DRB1 typing by HLAscan using data generated by NextGen were identical to those obtained using a Sanger sequencing–based method. We also applied HLAscan to a family dataset with various coverage depths generated on the Illumina HiSeq X-TEN platform. HLAscan identified allele types of HLA-A, −B, −C, −DQB1, and -DRB1 with 100% accuracy for sequences at ≥ 90× depth, and the overall accuracy was 96.9%. HLAscan, an alignment-based program that takes read distribution into account to determine true allele types, outperformed previously developed HLA typing tools. Therefore, HLAscan can be reliably applied for determination of HLA type across the whole-genome, exome, and target sequences.

Next-generation sequencing (NGS) technology has improved enough to discover mutations associated with genetic diseases. Our study evaluated the feasibility of targeted NGS as a primary screening tool to detect causal variants and subsequently predict genetic diseases. We performed parallel computations on 3.7-megabase-targeted regions to detect disease-causing mutations in 103 participants consisting of 81 patients and 22 controls. Data analysis of the participants took about 6 hours using local databases and 200 nodes of a supercomputer. All variants in the selected genes led on average to 3.6 putative diseases for each patient while variants restricted to disease-causing genes identified the correct disease. Notably, only 12% of predicted causal variants were recorded as causal mutations in public databases: 88% had no or insufficient records. In this study, most genetic diseases were caused by rare mutations and public records were inadequate. Most rare variants, however, were not associated with genetic diseases. These data implied that novel, rare variants should not be ignored but interpreted in conjunction with additional clinical data. This step is needed so appropriate advice can be given to primary doctors and parents, thus fulfilling the purpose of this method as a primary screen for rare genetic diseases.

Multiple epiphyseal dysplasia is a common skeletal dysplasia characterized by mild short stature, early-onset osteoarthritis mainly involving the hip and knee joints, and abnormally small and/or irregular epiphyses. Multiple epiphyseal dysplasia is clinically and genetically heterogeneous and six genes are associated with the phenotype of multiple epiphyseal dysplasia. A 12-year-old Korean boy presented with intermittent knee pain. His height was 144.6 cm (20th percentile) and family history was notable for early-onset osteoarthritis in his father. The proband’s x-rays revealed epiphyseal changes characteristic of multiple epiphyseal dysplasia associated with a collagen IX defect, with manifestations primarily restricted to the knees. Mutational analysis identified a novel c.104G > A substitution in exon 2 of COL9A3, resulting in p.Gly35Asp in the proband and his father. In silico analyses predicted the p.Gly35Asp amino acid change to be detelerious, and molecular dynamics simulation demonstrated a major structural change in the heterotrimeric collagen IX. So far, three COL9A3 mutations, have been reported. These three mutations are located at the splice donor or acceptor site of COL9A3 and cause skipping of exon 3, resulting in the deletion of 12 aminoacids in the COL3 domain of COL9A3. In contrast, the novel missense mutation identified in this two-generation family with multiple epiphyseal dysplasia is a missense mutation affecting the Gly residue of the Pro-Pro-Gly repeat sequence in the COL3 domain of collage IX, with accompanying major structural change of the collagen peptide.

Corneal dystrophies (CDs) are a diverse group of inherited disorders with a heterogeneous genetic background. Here, we report the identification of a novel ZNF143 heterozygous missense mutation in three individuals of the same family with clinical and pathological features that are consistent with endothelial CD. Ophthalmologic examination revealed diffuse corneal clouding and edema with decreased endothelial cell density. Pathological findings showed increased corneal thickness due to edema of basal epithelial cells and stroma, and abnormal metaplastic endothelium with stratified epithelium-like changes. Patients’ metaplastic corneal endothelial cells expressed predominantly cytokerain 7, cytokeratin 19, and E-cadherin. Although Sanger sequencing did not detect any mutation associated with endothelial CDs, whole exome sequencing identified the ZNF143 c.937G>C p.(Asp313His) mutation as a candidate gene for our patients’ endothelial CD. In-vitro functional studies demonstrated that mutant ZNF143 promoted the mesenchymal-to-epithelial transition; it upregulated the expression of genes associated with epithelialization in human corneal endothelial cells. Additionally, proinflammatory cytokine responsive genes were significantly enriched after mutant ZNF143 transfection, which may contribute to the severe phenotype of the three patients. These findings link a mutation in ZNF143 with endothelial CD for the first time.

Calling variants from next-generation sequencing (NGS) data or discovering discordant sequences between two NGS data sets is challenging. We developed a computer algorithm, ADIScan1, to call variants by comparing the fractions of allelic reads in a tester to the universal reference genome. We then created ADIScan2 by modifying the algorithm to directly compare two sets of NGS data and predict discordant sequences between two testers. ADIScan1 detected >99.7% of variants called by GATK with an additional 724 393 SNVs. ADIScan2 identified ∼500 candidates of discordant sequences in each of two pairs of the monozygotic twins. About 200 of these candidates were included in the ∼2800 predicted by VarScan2. We verified 66 true discordant sequences among the candidates that ADIScan2 and VarScan2 exclusively predicted. ADIScan2 detected many discordant sequences overlooked by VarScan2 and Mutect, which specialize in detecting low frequency mutations in genetically heterogeneous cancerous tissues. Numbers of verified sequences alone were >5 times more than expected based on recently estimated mutation rates from whole genome sequences. Estimated post-zygotic mutation rates were 1.68 × 10⁻⁷ in this study. ADIScan1 and 2 would complement existing tools in screening causative mutations of diverse genetic diseases and comparing two sets of genome sequences, respectively.