Citations

Abstract Explainable artificial intelligence aims to interpret how machine learning models make decisions, and many model explainers have been developed in the computer vision field. However, understanding of the applicability of these model explainers to biological data is still lacking. In this study, we comprehensively evaluated multiple explainers by interpreting pre-trained models for predicting tissue types from transcriptomic data and by identifying the top contributing genes from each sample with the greatest impacts on model prediction. To improve the reproducibility and interpretability of results generated by model explainers, we proposed a series of optimization strategies for each explainer on two different model architectures of multilayer perceptron (MLP) and convolutional neural network (CNN). We observed three groups of explainer and model architecture combinations with high reproducibility. Group II, which contains three model explainers on aggregated MLP models, identified top contributing genes in different tissues that exhibited tissue-specific manifestation and were potential cancer biomarkers. In summary, our work provides novel insights and guidance for exploring biological mechanisms using explainable machine learning models.

AbstractDefense systems are vital weapons for prokaryotes to resist heterologous DNA and survive from the constant invasion of viruses, and they are widely used in biochemistry investigation and antimicrobial drug research. So far, numerous types of defense systems have been discovered, but there is no comprehensive defense systems database to organize prokaryotic defense gene datasets. To fill this gap, we unveil the prokaryotic antiviral defense system (PADS) Arsenal (https://bigd.big.ac.cn/padsarsenal), a public database dedicated to gathering, storing, analyzing and visualizing prokaryotic defense gene datasets. The initial version of PADS Arsenal integrates 18 distinctive categories of defense system with the annotation of 6 600 264 genes retrieved from 63,701 genomes across 33 390 species of archaea and bacteria. PADS Arsenal provides various ways to retrieve defense systems related genes information and visualize them with multifarious function modes. Moreover, an online analysis pipeline is integrated into PADS Arsenal to facilitate annotation and evolutionary analysis of defense genes. PADS Arsenal can also visualize the dynamic variation information of defense genes from pan-genome analysis. Overall, PADS Arsenal is a state-of-the-art open comprehensive resource to accelerate the research of prokaryotic defense systems.

Functional FOXA1 targeting is mostly unique to each of the four human cancer cell lines HepG2, LNCaP, MCF7, and T47D.

Summary The transcription factor forkhead box protein A1 (FOXA1) plays a critical role in the proliferation of human breast cancer cells, particularly estrogen receptor alpha (ERα)‐positive luminal breast cancer cells. However, genetic studies of the requirement for Foxa1 in mammary tumor formation in mice have been hampered by the lack of a conditional gene ablation. We examined three mouse models of mammary‐specific ablation of Foxa1 in ductal epithelial cells to identify the best system for complete and mammary‐specific ablation of Foxa1. We found that MMTV‐Cre and MMTV‐rtTA;Tet‐On‐Cre led to partial deletion of Foxa1 and attenuated mammary duct formation, whereas Krt14‐Cre led to complete ablation of Foxa1 and abolished mammary duct formation, in Foxa1 loxP/loxP mice. These results demonstrate that Foxa1 is essential for mammary duct formation, and reveal a series of mouse models in which mammary expression of Foxa1 can be attenuated or completely blocked. Our study also suggests a potentially powerful model for complete ablation of Foxa1 in mammary epithelial cells using Krt14‐driven Cre expression in an inducible manner, such as Krt14‐rtTA;Tet‐On‐Cre . This model system will facilitate further in vivo functional studies of Foxa1 or other factors in mammary gland development and tumor formation and progression. genesis 54:277–285, 2016. © 2016 Wiley Periodicals, Inc.