Reducing variability of breast cancer subtype predictors by grounding deep learning models in prior knowledge.

Deep learning neural networks have improved performance in many cancer informatics problems, including breast cancer subtype classification. However, many networks experience underspecificationwheremultiplecombinationsofparametersachievesimilarperformance, bothin training and validation. Additionally, certain parameter combinations may perform poorly when the test distribution differs from the training distribution. Embedding prior knowledge from the literature may address this issue by boosting predictive models that provide crucial, in-depth information about a given disease. Breast cancer research provides a wealth of such knowledge, particularly in the form of subtype biomarkers and genetic signatures. In this study, we draw on past research on breast cancer subtype biomarkers, label propagation, and neural graph machines to present a novel methodology for embedding knowledge into machine learning systems. We embed prior knowledge into the loss function in the form of inter-subject distances derived from a well-known published breast cancer signature. Our results show that this methodology reduces predictor variability on state-of-the-art deep learning architectures and increases predictor consistency leading to improved interpretation. We find that pathway enrichment analysis is more consistent after embedding knowledge. This novel method applies to a broad range of existing studies and predictive models. Our method moves the traditional synthesis of predictive models from an arbitrary assignment of weights to genes toward a more biologically meaningful approach of incorporating knowledge.

Resources for the Comprehensive Discovery of Functional RNA Elements.

Transcriptome-wide maps of RNA binding protein (RBP)-RNA interactions by immunoprecipitation (IP)-based methods such as RNA IP (RIP) and crosslinking and IP (CLIP) are key starting points for evaluating the molecular roles of the thousands of human RBPs. A significant bottleneck to the application of these methods in diverse cell lines, tissues, and developmental stages is the availability of validated IP-quality antibodies. Using IP followed by immunoblot assays, we have developed a validated repository of 438 commercially available antibodies that interrogate 365 unique RBPs. In parallel, 362 short-hairpin RNA (shRNA) constructs against 276 unique RBPs were also used to confirm specificity of these antibodies. These antibodies can characterize subcellular RBP localization. With the burgeoning interest in the roles of RBPs in cancer, neurobiology, and development, these resources are invaluable to the broad scientific community. Detailed information about these resources is publicly available at the ENCODE portal (https://www.encodeproject.org/).

Integrative Genomics Implicates EGFR as a Downstream Mediator in NKX2-1 Amplified Non-Small Cell Lung Cancer.

NKX2-1, encoding a homeobox transcription factor, is amplified in approximately 15% of non-small cell lung cancers (NSCLC), where it is thought to drive cancer cell proliferation and survival. However, its mechanism of action remains largely unknown. To identify relevant downstream transcriptional targets, here we carried out a combined NKX2-1 transcriptome (NKX2-1 knockdown followed by RNAseq) and cistrome (NKX2-1 binding sites by ChIPseq) analysis in four NKX2-1-amplified human NSCLC cell lines. While NKX2-1 regulated genes differed among the four cell lines assayed, cell proliferation emerged as a common theme. Moreover, in 3 of the 4 cell lines, epidermal growth factor receptor (EGFR) was among the top NKX2-1 upregulated targets, which we confirmed at the protein level by western blot. Interestingly, EGFR knockdown led to upregulation of NKX2-1, suggesting a negative feedback loop. Consistent with this finding, combined knockdown of NKX2-1 and EGFR in NCI-H1819 lung cancer cells reduced cell proliferation (as well as MAP-kinase and PI3-kinase signaling) more than knockdown of either alone. Likewise, NKX2-1 knockdown enhanced the growth-inhibitory effect of the EGFR-inhibitor erlotinib. Taken together, our findings implicate EGFR as a downstream effector of NKX2-1 in NKX2-1 amplified NSCLC, with possible clinical implications, and provide a rich dataset for investigating additional mediators of NKX2-1 driven oncogenesis.

Doctors' age at domestic partnership and parenthood: cohort studies.

OBJECTIVE: To report on doctors' family formation. Design Cohort studies using structured questionnaires. Setting UK. Participants Doctors who qualified in 1988, 1993, 1996, 1999, 2000 and 2002 were followed up. MAIN OUTCOME MEASURES: Living with spouse or partner; and doctors' age when first child was born. RESULTS: The response to surveys including questions about domestic circumstances was 89.8% (20,717/23,077 doctors). The main outcomes - living with spouse or partner, and parenthood - varied according to age at qualification. Using the modal ages of 23-24 years at qualification, by the age of 24-25 (i.e. in their first year of medical work) a much smaller percentage of doctors than the general population was living with spouse or partner. By the age of 33, 75% of both women and men doctors were living with spouse or partner, compared with 68% of women and 61% of men aged 33 in the general population. By the age of 24-25, 2% of women doctors and 41% of women in the general population had a child; but women doctors caught up with the general population, in this respect, in their 30s. The specialty with the highest percentage of women doctors who, aged 35, had children was general practice (74%); the lowest was surgery (41%). CONCLUSIONS: Doctors are more likely than other people to live with a spouse or partner, and to have children, albeit typically at later ages. Differences between specialties in rates of motherhood may indicate sacrifice by some women of family in favour of career.

S phase-coupled E2f1 destruction ensures homeostasis in proliferating tissues.

Precise control of cell cycle regulators is critical for normal development and tissue homeostasis. E2F transcription factors are activated during G1 to drive the G1-S transition and are then inhibited during S phase by a variety of mechanisms. Here, we genetically manipulate the single Drosophila activator E2F (E2f1) to explore the developmental requirement for S phase-coupled E2F down-regulation. Expression of an E2f1 mutant that is not destroyed during S phase drives cell cycle progression and causes apoptosis. Interestingly, this apoptosis is not exclusively the result of inappropriate cell cycle progression, because a stable E2f1 mutant that cannot function as a transcription factor or drive cell cycle progression also triggers apoptosis. This observation suggests that the inappropriate presence of E2f1 protein during S phase can trigger apoptosis by mechanisms that are independent of E2F acting directly at target genes. The ability of S phase-stabilized E2f1 to trigger apoptosis requires an interaction between E2f1 and the Drosophila pRb homolog, Rbf1, and involves induction of the pro-apoptotic gene, hid. Simultaneously blocking E2f1 destruction during S phase and inhibiting the induction of apoptosis results in tissue overgrowth and lethality. We propose that inappropriate accumulation of E2f1 protein during S phase triggers the elimination of potentially hyperplastic cells via apoptosis in order to ensure normal development of rapidly proliferating tissues.

Loss of the histone pre-mRNA processing factor stem-loop binding protein in Drosophila causes genomic instability and impaired cellular proliferation.

BACKGROUND: Metazoan replication-dependent histone mRNAs terminate in a conserved stem-loop structure rather than a polyA tail. Formation of this unique mRNA 3' end requires Stem-loop Binding Protein (SLBP), which directly binds histone pre-mRNA and stimulates 3' end processing. The 3' end stem-loop is necessary for all aspects of histone mRNA metabolism, including replication coupling, but its importance to organism fitness and genome maintenance in vivo have not been characterized. METHODOLOGY/PRINCIPAL FINDINGS: In Drosophila, disruption of the Slbp gene prevents normal histone pre-mRNA processing and causes histone pre-mRNAs to utilize the canonical 3' end processing pathway, resulting in polyadenylated histone mRNAs that are no longer properly regulated. Here we show that Slbp mutants display genomic instability, including loss of heterozygosity (LOH), increased presence of chromosome breaks, tetraploidy, and changes in position effect variegation (PEV). During imaginal disc growth, Slbp mutant cells show defects in S phase and proliferate more slowly than control cells. CONCLUSIONS/SIGNIFICANCE: These data are consistent with a model in which changing the 3' end of histone mRNA disrupts normal replication-coupled histone mRNA biosynthesis and alters chromatin assembly, resulting in genomic instability, inhibition of cell proliferation, and impaired development.

Endoreplication: polyploidy with purpose.

A great many cell types are necessary for the myriad capabilities of complex, multicellular organisms. One interesting aspect of this diversity of cell type is that many cells in diploid organisms are polyploid. This is called endopolyploidy and arises from cell cycles that are often characterized as "variant," but in fact are widespread throughout nature. Endopolyploidy is essential for normal development and physiology in many different organisms. Here we review how both plants and animals use variations of the cell cycle, termed collectively as endoreplication, resulting in polyploid cells that support specific aspects of development. In addition, we discuss briefly how endoreplication occurs in response to certain physiological stresses, and how it may contribute to the development of cancer. Finally, we describe the molecular mechanisms that support the onset and progression of endoreplication.

Career preferences of graduate and non-graduate entrants to medical schools in the UK.

OBJECTIVE: Recent UK policy has been to increase substantially the number of graduate entrants to medical schools. Our aim was to study whether graduate and non-graduate entrants have different long-term career preferences. METHODS: We conducted postal questionnaire surveys of medical qualifiers from all UK medical schools in 1999, 2000 and 2002, surveyed 1 year after qualification, and qualifiers of 1999 and 2000, surveyed 3 years after qualification. RESULTS: By Year 3 after qualification, general practice was the choice of 33% of men graduate entrants and 21% of men non-graduates ( = 12.5, P < 0.001) and of 43% of women graduates and 38% of women non-graduates ( = 1.6, P = 0.2). Surgery was a much less popular choice for men graduate entrants than for men non-graduates; but similar percentages of women graduate and non-graduate entrants chose surgery. A lower percentage of graduate entrants than of non-graduates favoured paediatrics. Other differences between graduates and non-graduates were generally small. General practice was the preferred career for a much lower percentage of those who took an intercalated degree while at medical school, than of those who did not. CONCLUSIONS: Increasing graduate entry to medical school is likely to increase the percentage of doctors who want to become general practitioners, but only modestly so. It may also lead to a decline in the percentages choosing surgery and paediatrics. Otherwise, at least on the current criteria used for selecting students, increasing graduate entry will probably not make much difference to the percentage of newly qualified doctors seeking careers in different branches of practice.

Doctors' views of their first year of medical work and postgraduate training in the UK: questionnaire surveys.

OBJECTIVE: The first year of postgraduate work for newly qualified doctors in the UK, the pre-registration year, is spent working intensively in training posts under supervision. Our aim was to report the views of pre-registration doctors on these posts. DESIGN: Questionnaire survey. SUBJECTS: All medical graduates of 1999 and a 25% sample of graduates of 2000 from all UK medical schools. MAIN OUTCOME MEASURES: Doctors' views on the pre-registration house officer (PRHO) year, recorded as ratings in answers to questions and statements about the year. RESULTS: In reply to the question 'How much have you enjoyed the PRHO year overall?', rated on a scale from 0-10 (0 = no enjoyment; 10 = enjoyed it greatly), 44% of respondents (1341/3068) gave scores of 8-10; in all, 83.2% of respondents gave scores in the upper half of the scale (> or =6). However, there were criticisms of specific aspects of working conditions. Only a third agreed that their training during the year had been of a high standard. Posts in medicine were rated more highly than those in surgery for quality of training. Differences in views held by women and men junior doctors were few. However, where differences existed, women were slightly more positive about their work than men. CONCLUSION: Most graduates enjoyed the pre-registration year but there is still considerable scope for improvement in working conditions and training. Men and women gave similar responses, which suggests that later divergence in their career pathways is not attributable to different views formed about work in their pre-registration year.