Background: Accurate and robust pathological image analysis for colorectal cancer (CRC) diagnosis is time-consuming and knowledge-intensive, but is essential for CRC patients’ treatment. The current heavy workload of pathologists in clinics/hospitals may easily lead to unconscious misdiagnosis of CRC based on daily image analyses. Methods: Based on a state-of-the-art transfer-learned deep convolutional neural network in artificial intelligence (AI), we proposed a novel patch aggregation strategy for clinic CRC diagnosis using weakly labeled pathological whole-slide image (WSI) patches. This approach was trained and validated using an unprecedented and enormously large number of 170,099 patches, > 14,680 WSIs, from > 9631 subjects that covered diverse and representative clinical cases from multi-independent-sources across China, the USA, and Germany. Results: Our innovative AI tool consistently and nearly perfectly agreed with (average Kappa statistic 0.896) and even often better than most of the experienced expert pathologists when tested in diagnosing CRC WSIs from multicenters. The average area under the receiver operating characteristics curve (AUC) of AI was greater than that of the pathologists (0.988 vs 0.970) and achieved the best performance among the application of other AI methods to CRC diagnosis. Our AI-generated heatmap highlights the image regions of cancer tissue/cells. Conclusions: This first-ever generalizable AI system can handle large amounts of WSIs consistently and robustly without potential bias due to fatigue commonly experienced by clinical pathologists. It will drastically alleviate the heavy clinical burden of daily pathology diagnosis and improve the treatment for CRC patients. This tool is generalizable to other cancer diagnosis based on image recognition.

Machine-assisted pathological recognition has been focused on supervised learning (SL) that suffers from a significant annotation bottleneck. We propose a semi-supervised learning (SSL) method based on the mean teacher architecture using 13,111 whole slide images of colorectal cancer from 8803 subjects from 13 independent centers. SSL (~3150 labeled, ~40,950 unlabeled; ~6300 labeled, ~37,800 unlabeled patches) performs significantly better than the SL. No significant difference is found between SSL (~6300 labeled, ~37,800 unlabeled) and SL (~44,100 labeled) at patch-level diagnoses (area under the curve (AUC): 0.980 ± 0.014 vs. 0.987 ± 0.008, P value = 0.134) and patient-level diagnoses (AUC: 0.974 ± 0.013 vs. 0.980 ± 0.010, P value = 0.117), which is close to human pathologists (average AUC: 0.969). The evaluation on 15,000 lung and 294,912 lymph node images also confirm SSL can achieve similar performance as that of SL with massive annotations. SSL dramatically reduces the annotations, which has great potential to effectively build expert-level pathological artificial intelligence platforms in practice.

Osteoporosis is a prevalent bone metabolic disease and peripheral blood monocytes represent a major systemic cell type for bone metabolism. To identify the key dysfunctional pathways in osteoporosis, we performed pathway analyses on microarray data of monocytes from subjects with extremely high/low hip bone mineral density. We first performed a traditional pathway analysis for which different pathways were treated as independent. However, genes overlap among pathways will lead to “crosstalk” phenomenon, which may lead to false positive/negative results. Therefore, we applied correction techniques including a novel approach that considers the correlation among genes to adjust the crosstalk effects in the analysis. In traditional analysis, 10 pathways were found to be significantly associated with BMD variation. After correction for crosstalk effects, three of them remained significant. Moreover, the MAPK signaling pathway, which has been shown to be important for osteoclastogenesis, became significant only after the correction for crosstalk effects. We also identified a new module mainly consisting of genes present in mitochondria to be significant. In summary, we describe a novel method to correct the crosstalk effect in pathway analysis and found five key independent pathways involved in BMD regulation, which may provide a better understanding of biological functional networks in osteoporosis.

Accurate prognosis for cancer patients can provide critical information for optimizing treatment plans and improving life quality. Combining omics data and demographic/clinical information can offer a more comprehensive view of cancer prognosis than using omics or clinical data alone and can also reveal the underlying disease mechanisms at the molecular level. In this study, we developed and validated a deep learning framework to extract information from high-dimensional gene expression and miRNA expression data and conduct prognosis prediction for breast cancer and ovarian-cancer patients using multiple independent multi-omics datasets. Our model achieved significantly better prognosis prediction than the current machine learning and deep learning approaches in various settings. Moreover, an interpretation method was applied to tackle the “black-box” nature of deep neural networks and we identified features (i.e., genes, miRNA, demographic/clinical variables) that were important to distinguish predicted high- and low-risk patients. The significance of the identified features was partially supported by previous studies.

Whole genome sequencing studies are essential to obtain a comprehensive understanding of the vast pattern of human genomic variations. Here we report the results of a high-coverage whole genome sequencing study for 44 unrelated healthy Caucasian adults, each sequenced to over 50-fold coverage (averaging 65.8×). We identified approximately 11 million single nucleotide polymorphisms (SNPs), 2.8 million short insertions and deletions, and over 500,000 block substitutions. We showed that, although previous studies, including the 1000 Genomes Project Phase 1 study, have catalogued the vast majority of common SNPs, many of the low-frequency and rare variants remain undiscovered. For instance, approximately 1.4 million SNPs and 1.3 million short indels that we found were novel to both the dbSNP and the 1000 Genomes Project Phase 1 data sets, and the majority of which (∼96%) have a minor allele frequency less than 5%. On average, each individual genome carried ∼3.3 million SNPs and ∼492,000 indels/block substitutions, including approximately 179 variants that were predicted to cause loss of function of the gene products. Moreover, each individual genome carried an average of 44 such loss-of-function variants in a homozygous state, which would completely “knock out” the corresponding genes. Across all the 44 genomes, a total of 182 genes were “knocked-out” in at least one individual genome, among which 46 genes were “knocked out” in over 30% of our samples, suggesting that a number of genes are commonly “knocked-out” in general populations. Gene ontology analysis suggested that these commonly “knocked-out” genes are enriched in biological process related to antigen processing and immune response. Our results contribute towards a comprehensive characterization of human genomic variation, especially for less-common and rare variants, and provide an invaluable resource for future genetic studies of human variation and diseases.

Osteoporosis is characterized by low bone mineral density (BMD). The advancement of high-throughput technologies and integrative approaches provided an opportunity for deciphering the mechanisms underlying osteoporosis. Here, we generated genomic, transcriptomic, methylomic, and metabolomic datasets from 119 subjects with high (n = 61) and low (n = 58) BMDs. By adopting sparse multiple discriminative canonical correlation analysis, we identified an optimal multi-omics biomarker panel with 74 differentially expressed genes (DEGs), 75 differentially methylated CpG sites (DMCs), and 23 differential metabolic products (DMPs). By linking genetic data, we identified 199 targeted BMD-associated expression/methylation/metabolite quantitative trait loci (eQTLs/meQTLs/metaQTLs). The reconstructed networks/pathways showed extensive biomarker interactions, and a substantial proportion of these biomarkers were enriched in RANK/RANKL, MAPK/TGF-β, and WNT/β-catenin pathways and G-protein-coupled receptor, GTP-binding/GTPase, telomere/mitochondrial activities that are essential for bone metabolism. Five biomarkers (FADS2, ADRA2A, FMN1, RABL2A, SPRY1) revealed causal effects on BMD variation. Our study provided an innovative framework and insights into the pathogenesis of osteoporosis.