The human visual cortex enables visual perception through a cascade of hierarchical computations in cortical regions with distinct functionalities. Here, we introduce an AI-driven approach to discover the functional mapping of the visual cortex. We related human brain responses to scene images measured with functional MRI (fMRI) systematically to a diverse set of deep neural networks (DNNs) optimized to perform different scene perception tasks. We found a structured mapping between DNN tasks and brain regions along the ventral and dorsal visual streams. Low-level visual tasks mapped onto early brain regions, 3-dimensional scene perception tasks mapped onto the dorsal stream, and semantic tasks mapped onto the ventral stream. This mapping was of high fidelity, with more than 60% of the explainable variance in nine key regions being explained. Together, our results provide a novel functional mapping of the human visual cortex and demonstrate the power of the computational approach. <<<

Deep learning-based methods for deformable image registration are attractive alternatives to conventional registration methods because of their short registration times. However, these methods often fail to estimate larger displacements in complex deformation fields, for which a multi-resolution strategy is required. In this article, we propose to train neural networks progressively to address this problem. Instead of training a large convolutional neural network on the registration task all at once, we initially train smaller versions of the network on lower resolution versions of the images and deformation fields. During training, we progressively expand the network with additional layers that are trained on higher resolution data. We show that this way of training allows a network to learn larger displacements without sacrificing registration accuracy and that the resulting network is less sensitive to large misregistrations compared to training the full network all at once. We generate a large number of ground truth example data by applying random synthetic transformations to a training set of images, and test the network on the problem of intrapatient lung CT registration. We analyze the learned representations in the progressively growing network to assess how the progressive learning strategy influences training. Finally, we show that a progressive training procedure leads to improved registration accuracy when learning large and complex deformations. <<<

Jim Abraham, Amy B Heimberger, John Marshall, Elisabeth Heath, Joseph Drabick, Anthony Helmstetter, Joanne Xiu, Daniel Magee, Phillip Stafford, Chadi Nabhan, Sourabh Antani, Curtis Johnston, Matthew Oberley, Wolfgang Michael Korn, David Spetzler <<<

Cancer of Unknown Primary (CUP) occurs in 3-5% of patients when standard histological diagnostic tests are unable to determine the origin of metastatic cancer. Typically, a CUP diagnosis is treated empirically and has very poor outcomes, with median overall survival less than one year. Gene expression profiling alone has been used to identify the tissue of origin but struggles with low neoplastic percentage in metastatic sites which is where identification is often most needed. MI GPSai, a Genomic Prevalence Score, uses DNA sequencing and whole transcriptome data coupled with machine learning to aid in the diagnosis of cancer. The algorithm trained on genomic data from 34,352 cases and genomic and transcriptomic data from 23,137 cases and was validated on 19,555 cases. MI GPSai predicted the tumor type in the labeled data set with an accuracy of over 94% on 93% of cases while deliberating amongst 21 possible categories of cancer. When also considering the second highest prediction, the accuracy increases to 97%. Additionally, MI GPSai rendered a prediction for 71.7% of CUP cases. Pathologist evaluation of discrepancies between submitted diagnosis and MI GPSai predictions resulted in change of diagnosis in 41.3% of the time. MI GPSai provides clinically meaningful information in a large proportion of CUP cases and inclusion of MI GPSai in clinical routine could improve diagnostic fidelity. Moreover, all genomic markers essential for therapy selection are assessed in this assay, maximizing the clinical utility for patients within a single test. <<<

Ten years into the revival of deep networks and artificial intelligence, we propose a theoretical framework that sheds light on understanding deep networks within a bigger picture of Intelligence in general. We introduce two fundamental principles, Parsimony and Self-consistency, that address two fundamental questions regarding Intelligence: what to learn and how to learn, respectively. We believe the two principles are the cornerstones for the emergence of Intelligence, artificial or natural. While these two principles have rich classical roots, we argue that they can be stated anew in entirely measurable and computable ways. More specifically, the two principles lead to an effective and efficient computational framework, compressive closed-loop transcription, that unifies and explains the evolution of modern deep networks and many artificial intelligence practices. While we mainly use modeling of visual data as an example, we believe the two principles will unify understanding of broad families of autonomous intelligent systems and provide a framework for understanding the brain. <<<

OBJECTIVE: To study the effect of increasing endometrial thickness on live birth rates in fresh and frozen-thaw embryo transfer (FET) cycles.DESIGN: Retrospective cohort study.SETTING: National data from Autologous in vitro fertilization (IVF) embryo transfer and FET cycles in Canada from the Canadian Assisted Reproductive Technology Registry Plus (CARTR Plus) database for records between January 2013 and December 2019.PATIENTS: Thirty-three Canadians clinics participated in voluntary reporting of IVF and pregnancy outcomes to the Canadian Assisted Reproductive Technology Registry Plus database, and a total of 43,383 fresh and 53,377 frozen transfers were included.INTERVENTION(S): None.MAIN OUTCOME MEASURE(S): Clinical pregnancy, pregnancy loss, and live birth rates.RESULTS: In fresh IVF-embryo transfer cycles, increasing endometrial thickness is associated with significant increases in the mean number of oocytes retrieved, peak estradiol levels, number of usable embryos, clinical pregnancy rates, live birth rates, and mean term singleton birth weights, and a decrease in pregnancy loss rates. However, live birth rates plateau after 10-12 mm. In contrast, in FET cycles live birth rates plateau after the endometrium measures 7-10 mm. The improvement in live birth rates with increasing endometrial thickness was independent of patient age, timing of embryo transfer (e.g., cleavage stage vs. blastocyst stage), or the number of oocytes at retrieval.CONCLUSIONS: In cycles with a fresh embryo transfer, live birth rates increase significantly until an endometrial thickness of 10-12 mm, while in FET cycles live birth rates plateau after 7-10 mm. However, an endometrial thickness <6 mm was associated clearly with a dramatic reduction in live birth rates in fresh and frozen embryo transfer cycles. <<<

MOTIVATION: The growing use of next-generation sequencing and enlarged sequencing throughput require efficient short-read alignment, where seeding is one of the major performance bottlenecks. The key challenge in the seeding phase is searching for exact matches of substrings of short reads in the reference DNA sequence. Existing algorithms, however, present limitations in performance due to their frequent memory accesses.RESULTS: This article presents BWA-MEME, the first full-fledged short read alignment software that leverages learned indices for solving the exact match search problem for efficient seeding. BWA-MEME is a practical and efficient seeding algorithm based on a suffix array search algorithm that solves the challenges in utilizing learned indices for SMEM search which is extensively used in the seeding phase. Our evaluation shows that BWA-MEME achieves up to 3.45× speedup in seeding throughput over BWA-MEM2 by reducing the number of instructions by 4.60×, memory accesses by 8.77× and LLC misses by 2.21×, while ensuring the identical SAM output to BWA-MEM2.AVAILABILITY AND IMPLEMENTATION: The source code and test scripts are available for academic use at https://github.com/kaist-ina/BWA-MEME/.SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online. <<<

Single-cell Hi-C (scHi-C) can identify cell-to-cell variability of three-dimensional (3D) chromatin organization, but the sparseness of measured interactions poses an analysis challenge. Here we report Higashi, an algorithm based on hypergraph representation learning that can incorporate the latent correlations among single cells to enhance overall imputation of contact maps. Higashi outperforms existing methods for embedding and imputation of scHi-C data and is able to identify multiscale 3D genome features in single cells, such as compartmentalization and TAD-like domain boundaries, allowing refined delineation of their cell-to-cell variability. Moreover, Higashi can incorporate epigenomic signals jointly profiled in the same cell into the hypergraph representation learning framework, as compared to separate analysis of two modalities, leading to improved embeddings for single-nucleus methyl-3C data. In an scHi-C dataset from human prefrontal cortex, Higashi identifies connections between 3D genome features and cell-type-specific gene regulation. Higashi can also potentially be extended to analyze single-cell multiway chromatin interactions and other multimodal single-cell omics data. <<<

Abstract Background Efficient targeting to appropriate cell organelles is one of the bottlenecks for the production of recombinant proteins in plant systems. A common practice is to use the native secretory signal peptide of the heterologous protein to be produced. Though general features of secretion signals are conserved between plants and animals, the broad sequence variability among signal peptides suggests differing efficiency of signal peptide recognition. Results Aiming to improve secretion in moss bioreactors, we quantitatively compared the efficiency of two human signal peptides and six signals from recently isolated moss (Physcomitrella patens) proteins. We therefore used fusions of the different signals to heterologous reporter sequences for transient transfection of moss cells and measured the extra- and intracellular accumulation of the recombinant proteins rhVEGF and GST, respectively. Our data demonstrates an up to fivefold higher secretion efficiency with endogenous moss signals compared to the two utilised human signal peptides. Conclusion From the distribution of extra- and intracellular recombinant proteins, we suggest translational inhibition during the signal recognition particle-cycle (SRP-cycle) as the most probable of several possible explanations for the decreased extracellular accumulation with the human signals. In this work, we report on the supremacy of moss secretion signals over the utilised heterologous ones within the moss-bioreactor system. Though the molecular details of this effect remain to be elucidated, our results will contribute to the improvement of molecular farming systems. <<<

It is critical to discover novel biomarkers of lung cancer for establishing economical technology for diagnosis of lung cancer. Our study has suggested that the autofluorescence (AF) of the skin may become a novel biomarker of this type: First, development of lung cancer led to a significant increase in the skin's green AF in a mouse model of lung cancer; second, lung cancer patients had significantly higher skin's green AF at certain positions compared with healthy volunteers and pulmonary infection patients; and third, using the skin's green AF intensity at dorsal centremetacarpus as the variable, the areas under curve (AUC) for differentiating lung cancer patients and pulmonary infection patients and for differentiating lung cancer patients and healthy volunteers was 0.871 and 0.813, respectively. Collectively, our study has indicated that the skin's green AF at dorsal centremetacarpus may become a novel biomarker for establishing a ground-breaking diagnostic strategy for lung cancer. <<<

O Y Olivia Tse, Peiyong Jiang, Suk Hang Cheng, Wenlei Peng, Huimin Shang, John Wong, Stephen L Chan, Liona C Y Poon, Tak Y Leung, K C Allen Chan, Rossa W K Chiu, Y M Dennis Lo <<<

5-Methylcytosine (5mC) is an important type of epigenetic modification. Bisulfite sequencing (BS-seq) has limitations, such as severe DNA degradation. Using single molecule real-time sequencing, we developed a methodology to directly examine 5mC. This approach holistically examined kinetic signals of a DNA polymerase (including interpulse duration and pulse width) and sequence context for every nucleotide within a measurement window, termed the holistic kinetic (HK) model. The measurement window of each analyzed double-stranded DNA molecule comprised 21 nucleotides with a cytosine in a CpG site in the center. We used amplified DNA (unmethylated) and M.SssI-treated DNA (methylated) (M.SssI being a CpG methyltransferase) to train a convolutional neural network. The area under the curve for differentiating methylation states using such samples was up to 0.97. The sensitivity and specificity for genome-wide 5mC detection at single-base resolution reached 90% and 94%, respectively. The HK model was then tested on human-mouse hybrid fragments in which each member of the hybrid had a different methylation status. The model was also tested on human genomic DNA molecules extracted from various biological samples, such as buffy coat, placental, and tumoral tissues. The overall methylation levels deduced by the HK model were well correlated with those by BS-seq ( = 0.99; < 0.0001) and allowed the measurement of allele-specific methylation patterns in imprinted genes. Taken together, this methodology has provided a system for simultaneous genome-wide genetic and epigenetic analyses. <<<

The serotonin hypothesis of depression is still influential. We aimed to synthesise and evaluate evidence on whether depression is associated with lowered serotonin concentration or activity in a systematic umbrella review of the principal relevant areas of research. PubMed, EMBASE and PsycINFO were searched using terms appropriate to each area of research, from their inception until December 2020. Systematic reviews, meta-analyses and large data-set analyses in the following areas were identified: serotonin and serotonin metabolite, 5-HIAA, concentrations in body fluids; serotonin 5-HT receptor binding; serotonin transporter (SERT) levels measured by imaging or at post-mortem; tryptophan depletion studies; SERT gene associations and SERT gene-environment interactions. Studies of depression associated with physical conditions and specific subtypes of depression (e.g. bipolar depression) were excluded. Two independent reviewers extracted the data and assessed the quality of included studies using the AMSTAR-2, an adapted AMSTAR-2, or the STREGA for a large genetic study. The certainty of study results was assessed using a modified version of the GRADE. We did not synthesise results of individual meta-analyses because they included overlapping studies. The review was registered with PROSPERO (CRD42020207203). 17 studies were included: 12 systematic reviews and meta-analyses, 1 collaborative meta-analysis, 1 meta-analysis of large cohort studies, 1 systematic review and narrative synthesis, 1 genetic association study and 1 umbrella review. Quality of reviews was variable with some genetic studies of high quality. Two meta-analyses of overlapping studies examining the serotonin metabolite, 5-HIAA, showed no association with depression (largest n = 1002). One meta-analysis of cohort studies of plasma serotonin showed no relationship with depression, and evidence that lowered serotonin concentration was associated with antidepressant use (n = 1869). Two meta-analyses of overlapping studies examining the 5-HT receptor (largest n = 561), and three meta-analyses of overlapping studies examining SERT binding (largest n = 1845) showed weak and inconsistent evidence of reduced binding in some areas, which would be consistent with increased synaptic availability of serotonin in people with depression, if this was the original, causal abnormaly. However, effects of prior antidepressant use were not reliably excluded. One meta-analysis of tryptophan depletion studies found no effect in most healthy volunteers (n = 566), but weak evidence of an effect in those with a family history of depression (n = 75). Another systematic review (n = 342) and a sample of ten subsequent studies (n = 407) found no effect in volunteers. No systematic review of tryptophan depletion studies has been performed since 2007. The two largest and highest quality studies of the SERT gene, one genetic association study (n = 115,257) and one collaborative meta-analysis (n = 43,165), revealed no evidence of an association with depression, or of an interaction between genotype, stress and depression. The main areas of serotonin research provide no consistent evidence of there being an association between serotonin and depression, and no support for the hypothesis that depression is caused by lowered serotonin activity or concentrations. Some evidence was consistent with the possibility that long-term antidepressant use reduces serotonin concentration. <<<

BACKGROUND: Genome-wide RNA-sequencing technologies are increasingly critical to a wide variety of diagnostic and research applications. RNA-seq users often first enrich for mRNA, with the most popular enrichment method being poly(A) selection. In many applications it is well-known that poly(A) selection biases the view of the transcriptome by selecting for longer tailed mRNA species.RESULTS: Here, we show that poly(A) selection biases Oxford Nanopore direct RNA sequencing. As expected, poly(A) selection skews sequenced mRNAs toward longer poly(A) tail lengths. Interestingly, we identify a population of mRNAs (> 10% of genes' mRNAs) that are inconsistently captured by poly(A) selection due to highly variable poly(A) tails, and demonstrate this phenomenon in our hands and in published data. Importantly, we show poly(A) selection is dispensable for Oxford Nanopore's direct RNA-seq technique, and demonstrate successful library construction without poly(A) selection, with decreased input, and without loss of quality.CONCLUSIONS: Our work expands the utility of direct RNA-seq by validating the use of total RNA as input, and demonstrates important technical artifacts from poly(A) selection that inconsistently skew mRNA expression and poly(A) tail length measurements. <<<

Ruidong Wu , Fan Ding , Rui Wang , Rui Shen , Xiwen Zhang , Shitong Luo , Chenpeng Su , Zuofan Wu , Qi Xie , Bonnie Berger , Jianzhu Ma , Jian Peng <<<

Recent breakthroughs have used deep learning to exploit evolutionary information in multiple sequence alignments (MSAs) to accurately predict protein structures. However, MSAs of homologous proteins are not always available, such as with orphan proteins and fast-evolving proteins like antibodies, and a protein typically folds in a natural setting from its primary amino acid sequence into its three-dimensional structure, suggesting that evolutionary information and MSAs should not be necessary to predict a protein's folded form. Here, we introduce OmegaFold, the first computational method to successfully predict high-resolution protein structure from a single primary sequence alone. Using a new combination of a protein language model that allows us to make predictions from single sequences and a geometry-inspired transformer model trained on protein structures, OmegaFold outperforms RoseTTAFold and achieves similar prediction accuracy to AlphaFold2 on recently released structures. OmegaFold enables accurate predictions on orphan proteins that do not belong to any functionally characterized protein family and antibodies that tend to have noisy MSAs due to fast evolution. Our study fills a much-needed structure prediction gap and brings us a step closer to understanding protein folding in nature. <<<

Despite recent developments, it is hard to profile all multi-omics single-cell data modalities on the same cell. Thus, huge amounts of single-cell genomics data of unpaired observations on different cells are generated. We propose a method named UnpairReg for the regression analysis on unpaired observations to integrate single-cell multi-omics data. On real and simulated data, UnpairReg provides an accurate estimation of cell gene expression where only chromatin accessibility data is available. The cis-regulatory network inferred from UnpairReg is highly consistent with eQTL mapping. UnpairReg improves cell type identification accuracy by joint analysis of single-cell gene expression and chromatin accessibility data. <<<

We introduce Dual Coordinate VCF (DVCF), a file format that records genomic variants against two different reference genomes simultaneously and is fully compliant with the current VCF specification. As implemented in the Genozip platform, DVCF enables bioinformatics pipelines to seamlessly operate across two coordinate systems by leveraging the system most advantageous to each pipeline step, simplifying bioinformatics workflows and reducing file generation and associated data storage burden. Moreover, our benchmarking of Genozip DVCF shows that it produces more complete, less erroneous, and less biased translations across coordinate systems than two widely used alternative tools (i.e., LiftoverVcf and CrossMap). <<<

Ten years into the revival of deep networks and artificial intelligence, we propose a theoretical framework that sheds light on understanding deep networks within a bigger picture of Intelligence in general. We introduce two fundamental principles, Parsimony and Self-consistency, that address two fundamental questions regarding Intelligence: what to learn and how to learn, respectively. We believe the two principles are the cornerstones for the emergence of Intelligence, artificial or natural. While these two principles have rich classical roots, we argue that they can be stated anew in entirely measurable and computable ways. More specifically, the two principles lead to an effective and efficient computational framework, compressive closed-loop transcription, that unifies and explains the evolution of modern deep networks and many artificial intelligence practices. While we mainly use modeling of visual data as an example, we believe the two principles will unify understanding of broad families of autonomous intelligent systems and provide a framework for understanding the brain. <<<

Prader-Willi syndrome (PWS; MIM# 176270) is a neurodevelopmental disorder caused by the loss of expression of paternally imprinted genes within the PWS region located on 15q11.2. It is usually caused by either maternal uniparental disomy of chromosome 15 (UPD15) or 15q11.2 recurrent deletion(s). Here, we report a healthy carrier of a balanced X;15 translocation and her two daughters, both with the karyotype 45,X,der(X)t(X;15)(p22;q11.2),-15. Both daughters display symptoms consistent with haploinsufficiency of the SHOX gene and PWS. We explored the architecture of the derivative chromosomes and investigated effects on gene expression in patient-derived neural cells. First, a multiplex ligation-dependent probe amplification methylation assay was used to determine the methylation status of the PWS-region revealing maternal UPD15 in daughter 2, explaining her clinical symptoms. Next, short read whole genome sequencing and 10X genomics linked read sequencing was used to pinpoint the exact breakpoints of the translocation. Finally, we performed transcriptome sequencing on neuroepithelial stem cells from the mother and from daughter 1 and observed biallelic expression of genes in the PWS region (including SNRPN) in daughter 1. In summary, our multi-omics analysis highlights two different PWS mechanisms in one family and provide an example of how structural variation can affect imprinting through long-range interactions. <<<

Single-cell RNA sequencing (scRNA-seq) technology has contributed significantly to diverse research areas in biology, from cancer to development. Since scRNA-seq data is high-dimensional, a common strategy is to learn low dimensional latent representations better to understand overall structure in the data. In this work, we build upon scVI, a powerful deep generative model which can learn biologically meaningful latent representations, but which has limited explicit control of batch effects. Rather than prioritizing batch effect removal over conservation of biological variation, or vice versa, our goal is to provide a bird eye view of the trade-offs between these two conflicting objectives. Specifically, using the well established concept of Pareto front from economics and engineering, we seek to learn the entire trade-off curve between conservation of biological variation and removal of batch effects. A multi-objective optimisation technique known as Pareto multi-task learning (Pareto MTL) is used to obtain the Pareto front between conservation of biological variation and batch effect removal. Our results indicate Pareto MTL can obtain a better Pareto front than the naive scalarization approach typically encountered in the literature. In addition, we propose to measure batch effect by applying a neural-network based estimator called Mutual Information Neural Estimation (MINE) and show benefits over the more standard Maximum Mean Discrepancy (MMD) measure. The Pareto front between conservation of biological variation and batch effect removal is a valuable tool for researchers in computational biology. Our results demonstrate the efficacy of applying Pareto MTL to estimate the Pareto front in conjunction with applying MINE to measure the batch effect. <<<