Deep learning models for scanned document image classification and form understanding have made significant progress in the last few years. High accuracy can be achieved by a model with the help of copious amounts of labelled training data for closed-world classification. However, very little work has been done in the domain of fine-grained and head-tailed (class imbalance with some classes having high numbers of data points and some having a low number of data points) open-world classification for documents. Our proposed method achieves a better classification results than the baseline of the head-tail-novel/open dataset. Our techniques include separating the head-tail classes and transferring the knowledge from head data to the tail data. This transfer of knowledge also improves the capability of recognizing a novel category by 15

Progress in Automated Handwriting Recognition has been hampered by the lack of large training datasets. Nearly all research uses a set of small datasets that often cause models to overfit. We present CENSUS-HWR, a new dataset consisting of full English handwritten words in 1,812,014 gray scale images. A total of 1,865,134 handwritten texts from a vocabulary of 10,711 words in the English language are present in this collection. This dataset is intended to serve handwriting models as a benchmark for deep learning algorithms. This huge English handwriting recognition dataset has been extracted from the US 1930 and 1940 censuses taken by approximately 70,000 enumerators each year. The dataset and the trained model with their weights are freely available to download at https://censustree.org/data.html.

Many people suffer from social and emotional problems. These problems arc especially difficult for those who are blind or visually impaired, homeless, without family or community support, or incarcerated. We have developed several family history technology applications with an aim of helping these disadvantaged groups. We aim to show that family history technology has a high potential to improve social and emotional wellness for people, especially those who belong to these disadvantaged groups. In particular, we found that surveyed users of RelativeFinder reported that the application helped them feel more connected to their community and had a positive impact on their emotional well-being.

Entity extraction is an important step in document understanding. Higher accuracy entity extraction on fine-grained entities can be achieved by combining the utility of Named Entity Recognition (NER) and Relation Extraction (RE) models. In this paper, a cascading model is proposed that implements NER and Relation extraction. This model utilizes relations between entities to infer context-dependent fine-grain named entities in text corpora. The RE module runs independent of the NER module, which reduces error accumulation from sequential steps. This process improves on the fine-grained NER F1-score of existing state-of-the-art from .4753 to .8563 on our data, albeit on a strictly limited domain. This provides the potential for further applications in historical document processing. These applications will enable automated searching of historical documents, such as those used in economics research and family history.

Introgression is an important biological process affecting at least 10% of the extant species in the animal kingdom. Introgression significantly impacts inference of phylogenetic species relationships where a strictly binary tree model cannot adequately explain reticulate net-like species relationships. Here, we use phylogenomic approaches to understand patterns of introgression along the evolutionary history of a unique, nonmodel insect system: dragonflies and damselflies (Odonata). We demonstrate that introgression is a pervasive evolutionary force across various taxonomic levels within Odonata. In particular, we show that the morphologically “intermediate” species of Anisozygoptera (one of the three primary suborders within Odonata besides Zygoptera and Anisoptera), which retain phenotypic characteristics of the other two suborders, experienced high levels of introgression likely coming from zygopteran genomes. Additionally, we find evidence for multiple cases of deep inter-superfamilial ancestral introgression. [Gene flow; Odonata; phylogenomics; reticulate evolution.]

With the advent of deep neural networks, handwriting recognition systems have recently achieved remarkable performance. Unfortunately, to achieve high-quality results, these models require large amounts of labeled training data, which is difficult to obtain. Various methods have been proposed to reduce the volume of training data required. We propose a framework for fitting new handwriting recognition models that joins both active and transfer learning into a unified framework. Empirical results show that our method performs better than traditional active learning, transfer learning, and standard supervised training methods.

Historical documents contain huge amounts of information, but their handwritten nature has made it difficult and costly to digitize their information in an feffective way. Extraction of information from these sources has relied primarily on volunteer human indexers that manually transcribe the information into databases. Meanwhile, deep neural networks are becoming increasingly prevalent in most elds and have proven to be especially effective at image segmentation tasks and natural language processing. We take advantage of these strengths and create an end-to-end pipeline that automatically extracts and indexes causes of death from handwritten death records. This sequence of networks allows us to digitize information from death records more quickly, accurately, and cheaply than previously possible, providing access to future research opportunities previously untouched.

K-mers form the backbone of many bioinformatic algorithms. They are, however, difficult to store and use efficiently because the number of k-mers increases exponentially as k increases. Many algorithms exist for compressed storage of kmers but suffer from slow insert times or are probabilistic resulting in false-positive k-mers. Furthermore, k-mer libraries usually specialize in associating specific values with k-mers such as a color in colored de Bruijn Graphs or k-mer count. We present kcollections 1 , a compressed and parallel data structure designed for k-mers generated from whole, assembled genomes. Kcollections is available for C++ and provides set-and maplike structures as well as a k-mer counting data structure all of which utilize parallel operations designed using a MapReduce paradigm. Additionally, we provide basic Python bindings for rapid prototyping. Kcollections makes developing bioinformatic algorithms simpler by abstracting away the tedious task of storing k-mers.

Bacterial antibiotic resistance is becoming a significant health threat, and rapid identification of antibiotic-resistant bacteria is essential to save lives and reduce the spread of antibiotic resistance. This paper analyzes the ability of machine learning algorithms (MLAs) to process data from a novel spectroscopic diagnostic device to identify antibiotic-resistant genes and bacterial species by comparison to available bacterial DNA sequences. Simulation results show that the algorithms attain from 92% accuracy (for genes) up to 99% accuracy (for species). This novel approach identifies genes and species by optically reading the percentage of A, C, G, T bases in 1000s of short 10-base DNA oligomers instead of relying on conventional DNA sequencing in which the sequence of bases in long oligomers provides genetic information. The identification algorithms are robust in the presence of simulated random genetic mutations and simulated random experimental errors. Thus, these algorithms can be used to identify bacterial species, to reveal antibiotic resistance genes, and to perform other genomic analyses. Some MLAs evaluated here are shown to be better than others at accurate gene identification and avoidance of false negative identification of antibiotic resistance.

Comparing whole genomes and finding variation is an important and difficult bioinformatic task. We present the Polygraph, a data structure for reference-free, multiple whole genome alignment that can be used to identify genomic structural variation. This data structure is built from assembled genomes and preserves the genomic structure from the assembly. It avoids the “hairball” graph structure that can occur in other graph methods such as de Bruijn graphs. The Polygraph can easily be visualized and be used for identification of structural variants. We apply the Polygraph to Escherichia coli and Saccharomyces cerevisiae for finding Structural Variants.

In this article, we present GNUMAP (Genomic Next-generation Universal MAPper), a next-generation sequence read mapper, that utilizes a Full-text index in Minute Space (FM-index) with the Burrows-Wheeler Transform (BWT) as the reference genome data structure. Using the FM-index, GNUMAP is able to map reads with a dramatic decrease in memory usage while maintaining the same probabilistic mapping characteristics of previous versions.

Overbooking is frequently used to increase the revenue generated by a network infrastructure without incurring additional costs. If the overbooking factor is chosen appropriately, additional virtual circuits can be admitted without degrading quality of service for existing customers. Most implementations use a single factor to accept a linear fraction of traffic requests. If a piecewise linear approach is used in admissions, additional traffic can be accepted without causing proportional increases in loss rate and utilization. This additional accepted traffic can significantly improve the profit margin for network service providers.

Bacteria with resistance genes are becoming ever more common, and new methods of discovering antibiotics are being developed. One of these new methods involves researchers creating random peptides and testing their antimicrobial activity. Developing antibiotics from these peptides requires understanding which sequence motifs will be toxic to bacteria. To determine if the toxic peptides of a randomly-generated peptide library can be uniquely classified based solely on sequence motifs, we created the PepSeq Pipeline: a new software that utilizes a Random Forest algorithm to extract motifs from a peptide library. We found that this pipeline can accurately classify 56% of the toxic peptides in the peptide library using motifs extracted from the model. Testing on simulated data with less noise, we could classify up to 94% of the toxic peptides. The pipeline extracted significant toxic motifs in every library that was tested, but its ability to classify all toxic peptides depended on the number of motifs in the library. Once extracted, these motifs can be used both to understand the biology behind why certain peptides are toxic and to create novel antibiotics. The code and data used in this analysis can be found at https://github.com/tjense25/pep-seq-pipeline.

In today’s Internet, demand is increasing for guarantees of speed and efficiency. Current routers are very limited in the type and quantity of observed data they can provide, making it difficult for providers to maximize utilization without the risk of degraded throughput. This research uses statistical data currents provided by router vendors to estimate the impact of changes in network configuration on the probability of link overflow. This allows service providers to calculate in advance, the effect of grooming on a network, eliminating the conservative trial-and-error approach normally used. These predictions are made using Large Deviation Theory, which focuses on the tails of the distribution, giving a better estimate than average and peak values.

Comparing genomes to identify polymorphisms is a difficult task, especially beyond single nucleotide poly-morphisms. Polymorphism detection is important in disease association studies as well as in phylogenetic tree reconstruc-tion. We present a method for identifying polymorphisms in genomes by using a modified version de Bruijn graphs, data structures widely used in genome assembly from Next-Generation Sequencing. Using our method, we are able to identify polymorphisms that exist within a genome as well as well as see graph structures that form in the de Bruijn graph for particular types of polymorphisms (translocations, etc.).

We present kleuren, a novel assembly-free method to reconstruct phylogenetic trees using the Colored de Bruijn Graph. kleuren works by constructing the Colored de Bruijn Graph and then traversing it, finding bubble structures in the graph that provide phylogenetic signal. The bubbles are then aligned and concatenated to form a supermatrix, from which a phylogenetic tree is inferred. We introduce the algorithms that kleuren uses to accomplish this task, and show its performance on reconstructing the phylogenetic tree of 12 Drosophila species. kleuren reconstructed the established phylogenetic tree accurately, and is a viable tool for phylogenetic tree reconstruction using whole genome sequences. Software package available at: https://github.com/Colelyman/kleuren

Codon bias, the usage patterns of synonymous codons for encoding a protein sequence as nucleotides, is a biological phenomenon that is not fully understood. Several methods exist to represent the codon bias of an organism: codon adaptation index (CAI) [1], individual codon usage (ICU), hidden stop codons (HSC) [2] and codon context (CC) [3]. These methods are often employed in the optimization of heterologous gene expression to increase the accuracy and rate of translation. They, however, have many shortcomings as they dont take into account the local and global context of a gene. We present a method for modeling global and local codon bias through deep language models that is more robust than current methods by providing more contextual information and long-range dependencies.

BACKGROUND:Accurate detection of homologous relationships of biological sequences (DNA or amino acid) amongst organisms is an important and often difficult task that is essential to various evolutionary studies, ranging from building phylogenies to predicting functional gene annotations. There are many existing heuristic tools, most commonly based on bidirectional BLAST searches that are used to identify homologous genes and combine them into two fundamentally distinct classes: orthologs and paralogs. Due to only using heuristic filtering based on significance score cutoffs and having no cluster post-processing tools available, these methods can often produce multiple clusters constituting unrelated (non-homologous) sequences. Therefore sequencing data extracted from incomplete genome/transcriptome assemblies originated from low coverage sequencing or produced by de novo processes without a reference genome are susceptible to high false positive rates of homology detection.RESULTS:In this paper we develop biologically informative features that can be extracted from multiple sequence alignments of putative homologous genes (orthologs and paralogs) and further utilized in context of guided experimentation to verify false positive outcomes. We demonstrate that our machine learning method trained on both known homology clusters obtained from OrthoDB and randomly generated sequence alignments (non-homologs), successfully determines apparent false positives inferred by heuristic algorithms especially among proteomes recovered from low-coverage RNA-seq data. Almost ~42 % and ~25 % of predicted putative homologies by InParanoid and HaMStR respectively were classified as false positives on experimental data set.CONCLUSIONS:Our process increases the quality of output from other clustering algorithms by providing a novel post-processing method that is both fast and efficient at removing low quality clusters of putative homologous genes recovered by heuristic-based approaches.

Gene duplication plays a central role in adaptation to novel environments by providing new genetic material for functional divergence and evolution of biological complexity. Several evolutionary models have been proposed for gene duplication to explain how new gene copies are preserved by natural selection, but these models have rarely been tested using empirical data. Opsin proteins, when combined with a chromophore, form a photopigment that is responsible for the absorption of light, the first step in the phototransduction cascade. Adaptive gene duplications have occurred many times within the animal opsins' gene family, leading to novel wavelength sensitivities. Consequently, opsins are an attractive choice for the study of gene duplication evolutionary models. Odonata (dragonflies and damselflies) have the largest opsin repertoire of any insect currently known. Additionally, there is tremendous variation in opsin copy number between species, particularly in the long-wavelength-sensitive (LWS) class. Using comprehensive phylotranscriptomic and statistical approaches, we tested various evolutionary models of gene duplication. Our results suggest that both the blue-sensitive (BS) and LWS opsin classes were subjected to strong positive selection that greatly weakens after multiple duplication events, a pattern that is consistent with the permanent heterozygote model. Due to the immense interspecific variation and duplicability potential of opsin genes among odonates, they represent a unique model system to test hypotheses regarding opsin gene duplication and diversification at the molecular level.

BACKGROUND:Genome-wide association studies (GWAS) have effectively identified genetic factors for many diseases. Many diseases, including Alzheimer's disease (AD), have epistatic causes, requiring more sophisticated analyses to identify groups of variants which together affect phenotype.RESULTS:Based on the GWAS statistical model, we developed a multi-SNP GWAS analysis to identify pairs of variants whose common occurrence signaled the Alzheimer's disease phenotype.CONCLUSIONS:Despite not having sufficient data to demonstrate significance, our preliminary experimentation identified a high correlation between GRIA3 and HLA-DRB5 (an AD gene). GRIA3 has not been previously reported in association with AD, but is known to play a role in learning and memory.