Coreset selection methods have shown promise in reducing the training data size while maintaining model performance for data-efficient machine learning. However, many large real-world datasets suffer from unknown spurious correlations and hidden biases. Therefore, it is crucial to understand how such biases would affect downstream tasks via the selected coresets. In this work, we conduct the first comprehensive analysis of the implications of data selection on the bias levels of the selected coresets and the robustness of downstream models trained on them. We use an extensive experimental setting spanning ten different spurious correlations benchmarks, five score metrics to characterize sample importance/ difficulty, and five data selection policies across a broad range of coreset sizes to identify important patterns and derive insights. Thereby, we unravel a series of nontrivial nuances in well-known interactions between sample difficulty and bias alignment, as well as dataset bias and resultant model robustness. For example, we show that embedding-based sample characterizations run a comparatively lower risk of inadvertently exacerbating bias when used for selecting coresets compared to characterizations based on learning dynamics. Our analysis also reveals that lower bias levels achieved by coresets of difficult samples do not reliably guarantee downstream robustness. Most importantly, we show that special considerations need to be made when the coreset size is very small, since there is a unique risk of highly prototypical coresets reaching high average performance while obscuring their low group-robustness.

The task of out-of-distribution (OOD) detection is notoriously ill-defined. Earlier works focused on new-class detection, aiming to identify label-altering data distribution shifts, also known as "semantic shift." However, recent works argue for a focus on failure detection, expanding the OOD evaluation framework to account for label-preserving data distribution shifts, also known as "covariate shift.” Intriguingly, under this new framework, complex OOD detectors that were previously considered state-of-the-art now perform similarly to, or even worse than the simple maximum softmax probability baseline. This raises the question: what are the latest OOD detectors actually detecting? Deciphering the behavior of OOD detection algorithms requires evaluation datasets that decouples semantic shift and covariate shift. To aid our investigations, we present ImageNet-OOD, a clean semantic shift dataset that minimizes the interference of covariate shift. Through comprehensive experiments, we show that OOD detectors are more sensitive to covariate shift than to semantic shift, and the benefits of recent OOD detection algorithms on semantic shift detection is minimal. Our dataset and analyses provide important insights for guiding the design of future OOD detectors.

Dataset distillation has emerged as a strategy to overcome the hurdles associated with large datasets by learning a compact set of synthetic data that retains essential information from the original dataset. While distilled data can be used to train high performing models, little is understood about how the information is stored. In this study, we posit and answer three questions about the behavior, representativeness, and point-wise information content of distilled data. We reveal distilled data cannot serve as a substitute for real data during training outside the standard evaluation setting for dataset distillation. Additionally, the distillation process retains high task performance by compressing information related to the early training dynamics of real models. Finally, we provide an framework for interpreting distilled data and reveal that individual distilled data points contain meaningful semantic information. This investigation sheds light on the intricate nature of distilled data, providing a better understanding on how they can be effectively utilized.

When legged robots impact their environment, they undergo large changes in their velocities in a small amount of time. Measuring and applying feedback to these velocities is challenging, and is further complicated due to uncertainty in the impact model and impact timing. This work proposes a general framework for adapting feedback control during impact by projecting the control objectives to a subspace that is invariant to the impact event. The resultant controller is robust to uncertainties in the impact event while maintaining maximum control authority over the impact invariant subspace. We demonstrate the utility of the projection on a walking controller for a planar five-link-biped and on a jumping controller for a compliant 3D bipedal robot, Cassie. The effectiveness of our method is shown to translate well on hardware.

Efforts on cancer genomic study have revealed that mutations in cancer-driving oncogenes tend to mutually exclusive. To understand this tumorigenic mechanism and common cancer features, large-scale genomic study and molecular analyses across many cancer types are essential. We compiled somatic mutation and transcriptome profiles for a pan-cancer dataset that contains tumor tissue samples from 32 cancer types for over 10,000 patients. The genomic and molecular analysis was applied to four pan-cancer subgroups including pan-gastrointestinal, pan-gynecological, pan-kidney and pan-squamous, as well as the entire patient group. We identified seven genes, TP53, PIK3CA, PTEN, PIK3R1, LRP1B, CDH1, and TRRAP exhibited mutually exclusive mutation patterns across all cancer types (adjusted ). The mutations in these genes occurred in 55.9% of all cancer patients and over 70% of patients with pan-gastrointestinal, pan-gynecological and pan-squamous. Excepting LRP1B, the other six genes are included in the known cancer pathways. Notably, TP53, PIK3CA, and PIK3R1 are involved in over 20 cancer-related pathways. The co-expression network analysis suggested that LRP1B was associated with multiple genes that showed significant enrichment in Wnt signal pathway (adjusted p=0.0023) and ERBB signal pathway (adjusted p=0.0097). We further constructed a regulatory network to infer novel regulations of LRP1B in cancer. In addition, based on the mutations in the seven genes, we divided the patients into different groups for survival analysis. We found that the survival rate of the patient group with TP53 mutations was significantly shorter than in other groups (log rank p = 5.88e-10) in pan-gynecological. Our analysis systematically revealed the mutually exclusive mutation patterns of seven cancer genes across 32 cancer types. It helps us to better understanding the relations among different cancer types and pan-cancer subgroups, providing new insights into common mechanisms underlying diverse cancer types.

Diabetes mellitus and obesity are becoming some of the most serious public health challenges in the world. To help researchers more quickly reveal the complex relationships existing between diabetes mellitus, obesity, and related diseases in the literature, and give them an inspiration to search the effective treatments for these diseases, we propose a novel model named as representative latent Dirichlet allocation topic model (RLDA). We conducted the representation learning model on more than 337,000 pieces of diabetes and obesity related literature published in the recent decade. Then, an explicit analysis of the final result using a series of visualization tools to discover meaningful relations among diabetes mellitus, obesity, and other diseases was performed. In order to show the credibility of our discoveries, we used clinical reports, such as Standards of Medical Care in Diabetes, which were not used in our training data, to verify our results. Fortunately, a sufficient number of the reports were direct matches. With the help of our model, we achieved satisfactory results for diabetes mellitus and obesity. For example, we discovered that 22 other diseases are closely related to diabetes mellitus, 10 with obesity and 8 with both. In addition, the tumor, adolescent/child, inflammation, and hypertension will be the hottest research topics relating to diabetes and obesity in the near future. We believe that the representational learning model we have built can help biomedical researchers direct the focus and adjust the direction of their work.

Pulmonary adenocarcinoma in situ (AIS) is an intermediate subtype of lung adenocarcinoma that exhibits non-invasive growth patterns, but can develop into invasive. Almost 100% of AIS patients can be cured with complete resection. In contrast, the five-year survival rate for those diagnosed with invasive lung adenocarcinoma is only about 4%. In order to get a better understanding of adenocarcinoma and identify marks indicating its evaluation and progression, what needs to be done is a genome-wide evaluation of the disease. In this study, we used RNA-seq data from normal, AIS, and invasive lung cancer samples to identify gene module with differential gene expressions that represents the properties of AIS, distinct from normal and invasive tumor. Utilizing the differential expression patterns of protein-coding genes and long non-coding RNAs (lncRNAs) between AIS and other conditions, we obtained a small group of 72 AIS-specific genes consisting 41 protein-coding genes, 5 annotated lncRNAs, and 26 novel lncRNA transcripts that expressed specifically in AIS samples. We consider that twelve of the protein-coding genes are lung cancer driver genes with located driver somatic mutations. Moreover, these AIS-specific genes show capabilities (98% accuracy in an independent data set) for indicating early stage lung cancer and normal situations. These genes are determined based on cell-adhesion functioning, including angiogenesis and fibronectin, etc., that are highly related to cancer development. The comparison of AIS with normal and invasive tumor provides the gene list revealing the mechanisms that accounts for AIS progression to invasive cancer. Furthermore we identified important signatures contributing to the early diagnosis of lung cancer, providing auxiliary studies to our ongoing precision medicine research (http://americancse.org/events/csce2017/keynotes_lectures/yang_talk).

Thanks to recent advances in the field of genomics, it is now possible to create a comprehensive atlas of the basic units of life-cells. In this paper, we present a frame work for single cell genomics research which employs several new machine learning models such as convolutional neural networks, deep auto-encoder, recurrent neural networks etc. With these effective learning models on multi-source data, such as biomedical literatures and cell images, we can achieve novel cell types and functional gene sets.

Breast cancer is the most common type of invasive cancer in females. It accounts for 18.2% of all cancer deaths worldwide. Although somatic mutations play important roles in cancer development and prognosis, the outcome predictions are largely based on the expression of marker genes. We submit that developing an innovative prognostic model incorporating somatic mutations with gene expression can improve survival prediction of cancer. We studied the whole genome-wide gene expression and somatic mutations of 1091 breast invasive carcinoma cases from The Cancer Genome Atlas (TCGA). We identified expression of 118 genes that could be used to build the predictor for breast cancer survival risks (log rank p <; 0.0001 and c-index=0.63627122). Multiple breast cancer survival-analysis-related genes are found in this gene set, such as FOXR2, FOXD1, MTNR1B, SDC1, PF4, IGF2BP1, ZIC3, OXT2, and PID1. We then selected between different survival-risk groups of 2000 mutated genes with different mutation rates. We applied enrichment analysis to the mutated gene list and identified 25 functional annotations, 15 gene ontology, and 14 gene pathway enriched terms that were related the cancer outcomes. We built the novel predictor and our results showed that combining the different features helps improved performance of survival prediction (c-index = 0. 64033769). Thus our model can be used to facilitate the advancement of our going precision medicine research (http://americancse.org/events/csce2017/keynotes_lectures/yang_talk).

The completeness of gene expression data is essential to many gene expression data analysis issues. In this paper, inspired by the idea of semi-supervised learning with tri-training, a hybrid iterative imputation method called tri-imputation is proposed to estimate the missing values in gene expression data. In detail, in each round of tri-imputation, any two imputation methods are collaborating with each other to firstly estimate an initial imputation value, and then to be applied to the rest imputation method for providing different available information. Finally, all these three results are combined with their respective pre-trained confidence values. Experimental results on real microarray matrices indicate that tri-imputation achieves more accurate estimation for missing values in terms of the lowest normalized root-mean-square error.

Efficacy prediction is an inseparable part of TCM. We firstly analyze the correlation between indicators and efficacy, and max blood-drug concentration(Cmax) is chosen as the target to reflect the efficacy of drugs. Then we apply linear regression(LR), support vector regression(SVR) as well as artificial neural networks(ANNs) to predict the efficacy of Wuji pills. The results of the leave-one-out method show that SVR performs better than other methods for label Cmax, and appears to be a good method for this task. In order to find the relationship between each component of Wuji Pills, several visualization methods are adopted to deal with this problem. The web server of prediction is available at http://data.jindengtai.cn/#/case/drug for public usage.

The antioxidant activity of green tea polyphenol epigallocatechin-3-gallate (EGCG) has been found to be critical in inhibiting carcinogenesis. In our previous study, we identified a set of protein coding genes and microRNAs whose expressions were significantly modulated in response to the EGCG treatment in tobacco carcinogen-induced lung adenocarcinoma in A/J mice. In this study, we further conducted some statistical analysis on our microarray data and employed The Cancer Genome Atlas (TCGA) lung adenocarcinoma datasets as additional control. We postulated that if a gene mediates EGCG's cancer inhibition, its expression level change caused by EGCG should be opposite to what occurred in the carcinogenesis. With this assumption, we identified Myb and Peg3 as the primary genes involved in the cancer inhibitory activities of EGCG.

BACKGROUND: Combining information from different studies is an important and useful practice in bioinformatics, including genome-wide association study, rare variant data analysis and other set-based analyses. Many statistical methods have been proposed to combine p-values from independent studies. However, it is known that there is no uniformly most powerful test under all conditions; therefore, finding a powerful test in specific situation is important and desirable. RESULTS: In this paper, we propose a new statistical approach to combining p-values based on gamma distribution, which uses the inverse of the p-value as the shape parameter in the gamma distribution. CONCLUSIONS: Simulation study and real data application demonstrate that the proposed method has good performance under some situations.

BACKGROUND: Diabetes mellitus of type 2 (T2D), also known as noninsulin-dependent diabetes mellitus (NIDDM) or adult-onset diabetes, is a common disease. It is estimated that more than 300 million people worldwide suffer from T2D. In this study, we investigated the T2D, pre-diabetic and healthy human (no diabetes) bloodstream samples using genomic, genealogical, and phonemic information. We identified differentially expressed genes and pathways. The study has provided deeper insights into the development of T2D, and provided useful information for further effective prevention and treatment of the disease. RESULTS: A total of 142 bloodstream samples were collected, including 47 healthy humans, 22 pre-diabetic and 73 T2D patients. Whole genome scale gene expression profiles were obtained using the Agilent Oligo chips that contain over 20,000 human genes. We identified 79 significantly differentially expressed genes that have fold change ≥ 2. We mapped those genes and pinpointed locations of those genes on human chromosomes. Amongst them, 3 genes were not mapped well on the human genome, but the rest of 76 differentially expressed genes were well mapped on the human genome. We found that most abundant differentially expressed genes are on chromosome one, which contains 9 of those genes, followed by chromosome two that contains 7 of the 76 differentially expressed genes. We performed gene ontology (GO) functional analysis of those 79 differentially expressed genes and found that genes involve in the regulation of cell proliferation were among most common pathways related to T2D. The expression of the 79 genes was combined with clinical information that includes age, sex, and race to construct an optimal discriminant model. The overall performance of the model reached 95.1% accuracy, with 91.5% accuracy on identifying healthy humans, 100% accuracy on pre-diabetic patients and 95.9% accuract on T2D patients. The higher performance on identifying pre-diabetic patients was resulted from more significant changes of gene expressions among this particular group of humans, which implicated that patients were having profound genetic changes towards disease development. CONCLUSION: Differentially expressed genes were distributed across chromosomes, and are more abundant on chromosomes 1 and 2 than the rest of the human genome. We found that regulation of cell proliferation actually plays an important role in the T2D disease development. The predictive model developed in this study has utilized the 79 significant genes in combination with age, sex, and racial information to distinguish pre-diabetic, T2D, and healthy humans. The study not only has provided deeper understanding of the disease molecular mechanisms but also useful information for pathway analysis and effective drug target identification.

Advances of high-throughput technologies have rapidly produced more and more data from DNAs and RNAs to proteins, especially large volumes of genome-scale data. However, connection of the genomic information to cellular functions and biological behaviours relies on the development of effective approaches at higher systems level. In particular, advances in RNA-Seq technology has helped the studies of transcriptome, RNA expressed from the genome, while systems biology on the other hand provides more comprehensive pictures, from which genes and proteins actively interact to lead to cellular behaviours and physiological phenotypes. As biological interactions mediate many biological processes that are essential for cellular function or disease development, it is important to systematically identify genomic information including genetic mutations from GWAS (genome-wide association study), differentially expressed genes, bidirectional promoters, intrinsic disordered proteins (IDP) and protein interactions to gain deep insights into the underlying mechanisms of gene regulations and networks. Furthermore, bidirectional promoters can co-regulate many biological pathways, where the roles of bidirectional promoters can be studied systematically for identifying co-regulating genes at interactive network level. Combining information from different but related studies can ultimately help revealing the landscape of molecular mechanisms underlying complex diseases such as cancer.

BACKGROUND: Kidney Renal Clear Cell Carcinoma (KIRC) is one of fatal genitourinary diseases and accounts for most malignant kidney tumours. KIRC has been shown resistance to radiotherapy and chemotherapy. Like many types of cancers, there is no curative treatment for metastatic KIRC. Using advanced sequencing technologies, The Cancer Genome Atlas (TCGA) project of NIH/NCI-NHGRI has produced large-scale sequencing data, which provide unprecedented opportunities to reveal new molecular mechanisms of cancer. We combined differentially expressed genes, pathways and network analyses to gain new insights into the underlying molecular mechanisms of the disease development. RESULTS: Followed by the experimental design for obtaining significant genes and pathways, comprehensive analysis of 537 KIRC patients' sequencing data provided by TCGA was performed. Differentially expressed genes were obtained from the RNA-Seq data. Pathway and network analyses were performed. We identified 186 differentially expressed genes with significant p-value and large fold changes (P < 0.01, |log(FC)| > 5). The study not only confirmed a number of identified differentially expressed genes in literature reports, but also provided new findings. We performed hierarchical clustering analysis utilizing the whole genome-wide gene expressions and differentially expressed genes that were identified in this study. We revealed distinct groups of differentially expressed genes that can aid to the identification of subtypes of the cancer. The hierarchical clustering analysis based on gene expression profile and differentially expressed genes suggested four subtypes of the cancer. We found enriched distinct Gene Ontology (GO) terms associated with these groups of genes. Based on these findings, we built a support vector machine based supervised-learning classifier to predict unknown samples, and the classifier achieved high accuracy and robust classification results. In addition, we identified a number of pathways (P < 0.04) that were significantly influenced by the disease. We found that some of the identified pathways have been implicated in cancers from literatures, while others have not been reported in the cancer before. The network analysis leads to the identification of significantly disrupted pathways and associated genes involved in the disease development. Furthermore, this study can provide a viable alternative in identifying effective drug targets. CONCLUSIONS: Our study identified a set of differentially expressed genes and pathways in kidney renal clear cell carcinoma, and represents a comprehensive computational approach to analysis large-scale next-generation sequencing data. The pathway and network analyses suggested that information from distinctly expressed genes can be utilized in the identification of aberrant upstream regulators. Identification of distinctly expressed genes and altered pathways are important in effective biomarker identification for early cancer diagnosis and treatment planning. Combining differentially expressed genes with pathway and network analyses using intelligent computational approaches provide an unprecedented opportunity to identify upstream disease causal genes and effective drug targets.