We present an unsupervised learning pipeline for dense depth, optical flow and egomotion estimation for autonomous driving applications, using the event-based output of the Dynamic Vision Sensor (DVS) as input. The backbone of our pipeline is a bioinspired encoder-decoder neural network architecture - ECN. To train the pipeline, we introduce a covariance normalization technique which resembles the lateral inhibition mechanism found in animal neural systems.Our work is the first monocular pipeline that generates dense depth and optical flow from sparse event data only, and is able to transfer from day to night scenes without any additional training. The network works in self-supervised mode and has just 150k parameters. We evaluate our pipeline on the MVSEC self driving dataset and present results for depth, optical flow and and egomotion estimation. Thanks to the efficient design, we are able to achieve inference rates of 300 FPS on a single Nvidia 1080Ti GPU. Our experiments demonstrate significant improvements upon works that used deep learning on event data, as well as the ability to perform well during both day and night.

MOTIVATION: Second-generation sequencing technologies produce high coverage of the genome by short reads at a low cost, which has prompted development of new assembly methods. In particular, multiple algorithms based on de Bruijn graphs have been shown to be effective for the assembly problem. In this article, we describe a new hybrid approach that has the computational efficiency of de Bruijn graph methods and the flexibility of overlap-based assembly strategies, and which allows variable read lengths while tolerating a significant level of sequencing error. Our method transforms large numbers of paired-end reads into a much smaller number of longer 'super-reads'. The use of super-reads allows us to assemble combinations of Illumina reads of differing lengths together with longer reads from 454 and Sanger sequencing technologies, making it one of the few assemblers capable of handling such mixtures. We call our system the Maryland Super-Read Celera Assembler (abbreviated MaSuRCA and pronounced 'mazurka'). RESULTS: We evaluate the performance of MaSuRCA against two of the most widely used assemblers for Illumina data, Allpaths-LG and SOAPdenovo2, on two datasets from organisms for which high-quality assemblies are available: the bacterium Rhodobacter sphaeroides and chromosome 16 of the mouse genome. We show that MaSuRCA performs on par or better than Allpaths-LG and significantly better than SOAPdenovo on these data, when evaluated against the finished sequence. We then show that MaSuRCA can significantly improve its assemblies when the original data are augmented with long reads. AVAILABILITY: MaSuRCA is available as open-source code at ftp://ftp.genome.umd.edu/pub/MaSuRCA/. Previous (pre-publication) releases have been publicly available for over a year. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Sequences produced by automated Sanger sequencing machines frequently contain fragments of the cloning vector on their ends. Software tools currently available for identifying and removing the vector sequence require knowledge of the vector sequence, specific splice sites and any adapter sequences used in the experiment-information often omitted from public databases. Furthermore, the clipping coordinates themselves are missing or incorrectly reported. As an example, within the approximately 1.24 billion shotgun sequences deposited in the NCBI Trace Archive, as many as approximately 735 million (approximately 60%) lack vector clipping information. Correct clipping information is essential to scientists attempting to validate, improve and even finish the increasingly large number of genomes released at a 'draft' quality level. RESULTS: We present here Figaro, a novel software tool for identifying and removing the vector from raw sequence data without prior knowledge of the vector sequence. The vector sequence is automatically inferred by analyzing the frequency of occurrence of short oligo-nucleotides using Poisson statistics. We show that Figaro achieves 99.98% sensitivity when tested on approximately 1.5 million shotgun reads from Drosophila pseudoobscura. We further explore the impact of accurate vector trimming on the quality of whole-genome assemblies by re-assembling two bacterial genomes from shotgun sequences deposited in the Trace Archive. Designed as a module in large computational pipelines, Figaro is fast, lightweight and flexible. AVAILABILITY: Figaro is released under an open-source license through the AMOS package (http://amos.sourceforge.net/Figaro).

MOTIVATION: Many genomes are sequenced by a collaboration of several centers, and then each center produces an assembly using their own assembly software. The collaborators then pick the draft assembly that they judge to be the best and the information contained in the other assemblies is usually not used. METHODS: We have developed a technique that we call assembly reconciliation that can merge draft genome assemblies. It takes one draft assembly, detects apparent errors, and, when possible, patches the problem areas using pieces from alternative draft assemblies. It also closes gaps in places where one of the alternative assemblies has spanned the gap correctly. RESULTS: Using the Assembly Reconciliation technique, we produced reconciled assemblies of six Drosophila species in collaboration with Agencourt Bioscience and The J. Craig Venter Institute. These assemblies are now the official (CAF1) assemblies used for analysis. We also produced a reconciled assembly of Rhesus Macaque genome, and this assembly is available from our website http://www.genome.umd.edu. AVAILABILITY: The reconciliation software is available for download from http://www.genome.umd.edu/software.htm

With hundreds of genomes now in GenBank, researchers might be forgiven for assuming that genome sequence data are correct, at least at a large scale. Certainly there might be errors at some small rate, perhaps 1 in 50 000 or 100 000 bases (Schmutz et al., 2004; Read et al., 2002), but at a large scale these genomes are put together correctly, are not they? Well, not always. We have been looking at the assemblies of large genomes for several years now, and for every ‘draft’ genome we look at, we find hundreds—and sometimes thousands—of mis-assemblies. These include regions where a genome is incorrectly re-arranged as well as places where large chunks of DNA sequence are simply deleted and the surrounding sequences just crunched together. The source of most mis-assemblies is, as it has always been, repeats. Genomes vary in their repeat content, but we have learned that large genomes are filled with repeats of all shapes and sizes. To illustrate how these repeats result in sequences being ‘lost’ by an assembler, consider the situation in Figure 1. Assemblies can collapse around repetitive sequences. R1 and R2, in yellow, represent near-identical copies of the same DNA sequence. In the figure, we see that the genome has two copies, R1 and R2, of a sequence that lie near one another, separated by a unique region shown in red. If R1 and R2 are long enough, then the assembler will not have any individual sequences (‘reads’) containing the entire repeat and its unique flanking sequences (the green and blue regions). The result will be that the genome assembly looks like the lower half of the figure, with a contiguous stretch of DNA (a contig) that has just one copy of the repeat, incorrectly jamming together the blue and green regions, and the red region will have no place to go. If this seems like a made-up example, it is not: we have observed that even the best assemblers today make exactly this mistake when assembling the Drosophila species currently being sequenced. Compressions such as this can easily total 1% or more of the genome, and the ‘orphan’ regions can be quite long, 5000–10 000 bp or more. And we would note that Drosophila is not a particularly difficult genome as compared with many others currently under way. To those who might think (or argue) that the assembler they are using is not prone to such errors, we can only reply that we have seen these types of errors in all the major assemblers in use today (e.g. Arachne (Batzoglou et al., 2002; Jaffe et al., 2003), Celera Assembler (Myers et al., 2000), Jazz (Aparicio et al., 2002), Phusion (Mullikin and Ning, 2003), PCAP (Huang et al., 2003) and Atlas (Havlak et al., 2004)), in some cases after running the assemblers ourselves and in other cases after carefully examining the results of assemblies created by others. We have developed software for improving assemblies that can detect at least some situations like the one shown above, although there is still no automated way of fixing these problems. However, the problem is often made much more difficult by the diploid nature of most large genomes, particularly the many mammalian genomes currently being sequenced by the NIH. The problem is this: the two copies of a chromosome are always slightly divergent, and this has led assembly groups (including ours) to develop methods for separating the two haplotypes from one another. But wherever there are tandem repeats in two or more copies, it can become extremely difficult to distinguish an incorrectly collapsed repeat (including situations such as that shown in Fig. 1) from true polymorphisms between the haplotypes. A tremendous amount of genome analysis is built upon the framework of the DNA sequence itself: not only are genes and regulatory sites anchored in the sequence, but analyses of synteny, duplications and evolutionary relationships among species all depend on having the correct structure of the genome. We need to devote more effort to making sure the basis for all these analyses does not turn out to be a house of cards. Our group has created a website (Author Webpage) for depositing reference assemblies: genomes for which the sequence is finished, and for which we can demonstrate how all the original data map to that finished sequence. The site also distinguishes the original whole-genome shotgun reads from any additional finishing reads. This small set of genomes, which thus far only includes bacteria, should be just the beginning: all assemblies need to be available so that others can check them and, if necessary, correct them. Fortunately, NCBI has created a much larger resource to capture both draft and finished assemblies, the Assembly Archive (Salzberg et al., 2004). This archive captures the complete information about how a set of raw sequences maps to a genome assembly, whether that assembly is ‘draft’ or ‘finished’. After spending fifteen years and hundreds of millions of dollars on the human genome, the community has a near-complete draft sequence, but the evidence for that sequence—the underlying raw data and the assembly itself—is, amazingly, not available. Indeed, many of the original assemblies of parts of the human genome were done in the mid- and late-1990s, and are now lost. We can only hope that future genomes would not be needlessly lost now that there is a place to deposit them. Are we arguing that all genomes should be finished? Actually, finishing does not necessarily address this problem at all. Finishing efforts are usually directed at closing gaps, not at fixing mis-assemblies, and therefore ‘finished’ genomes are very likely to contain errors of the type we are discussing. A better term for such genomes is ‘closed’: gaps are closed but sequence is not confirmed. We strongly suspect that many of the already-published finished genomes in GenBank today contain assembly errors. Clearly we also need new, well-defined methods for comparing assemblies. The most popular metrics right now all seem to emphasize size: size of contigs, size of scaffolds, and especially N50 sizes. (The N50 size is computed by sorting all contigs from largest to smallest and by determining the minimum set of contigs whose sizes total 50% of the entire genome. The N50 size is the smallest contig in that set.) The standard of judging assembly quality by size of contigs is questionable. Large contigs may simply reflect overly aggressive joining of contigs, thereby creating larger contigs with mis-assemblies. As a consequence, genome scientists who are not experts at assembly can be completely misled by statistics about contig sizes, and as a result might prefer the ‘larger’ but incorrect assembly when given a choice. We need to start capturing assemblies and looking at them with a more skeptical eye. This need has become even greater in the face of a growing number of ‘draft’ assemblies, many of which will never be finished. Before launching lengthy projects based on these genomes, we need to be confident that they are assembled correctly. The bioinformatics community should take the lead in this effort, by developing standards for quality control and by devoting more time and energy to careful evaluations of genome assemblies.

MOTIVATION: Comparison of nucleic acid and protein sequences is a fundamental tool of modern bioinformatics. A dominant method of such string matching is the 'seed-and-extend' approach, in which occurrences of short subsequences called 'seeds' are used to search for potentially longer matches in a large database of sequences. Each such potential match is then checked to see if it extends beyond the seed. To be effective, the seed-and-extend approach needs to catalogue seeds from virtually every substring in the database of search strings. Projects such as mammalian genome assemblies and large-scale protein matching, however, have such large sequence databases that the resulting list of seeds cannot be stored in RAM on a single computer. This significantly slows the matching process. RESULTS: We present a simple and elegant method in which only a small fraction of seeds, called 'minimizers', needs to be stored. Using minimizers can speed up string-matching computations by a large factor while missing only a small fraction of the matches found using all seeds.