Controlled execution of dynamic motions in quadrupedal robots, especially those with articulated soft bodies, presents a unique set of challenges that traditional methods struggle to address efficiently. In this study, we tackle these issues by relying on a simple yet effective two-stage learning framework to generate dynamic motions for quadrupedal robots. First, a gradient-free evolution strategy is employed to discover simply represented control policies, eliminating the need for a predefined reference motion. Then, we refine these policies using deep reinforcement learning. Our approach enables the acquisition of complex motions like pronking and back-flipping, effectively from scratch. Additionally, our method simplifies the traditionally labour-intensive task of reward shaping, boosting the efficiency of the learning process. Importantly, our framework proves particularly effective for articulated soft quadrupeds, whose inherent compliance and adaptability make them ideal for dynamic tasks but also introduce unique control challenges.

BACKGROUND: Bisulfite treatment of DNA followed by sequencing (BS-seq) has become a standard technique in epigenetic studies, providing researchers with tools for generating single-base resolution maps of whole methylomes. Aligning bisulfite-treated reads, however, is a computationally difficult task: bisulfite treatment decreases the (lexical) complexity of low-methylated genomic regions, and C-to-T mismatches may reflect cytosine unmethylation rather than SNPs or sequencing errors. Further challenges arise both during and after the alignment phase: data structures used by the aligner should be fast and should fit into main memory, and the methylation-caller output should be somehow compressed, due to its significant size. METHODS: As far as data structures employed to align bisulfite-treated reads are concerned, solutions proposed in the literature can be roughly grouped into two main categories: those storing pointers at each text position (e.g. hash tables, suffix trees/arrays), and those using the information-theoretic minimum number of bits (e.g. FM indexes and compressed suffix arrays). The former are fast and memory consuming. The latter are much slower and light. In this paper, we try to close this gap proposing a data structure for aligning bisulfite-treated reads which is at the same time fast, light, and very accurate. We reach this objective by combining a recent theoretical result on succinct hashing with a bisulfite-aware hash function. Furthermore, the new versions of the tools implementing our ideas|the aligner ERNE-BS5 2 and the caller ERNE-METH 2|have been extended with increased downstream compatibility (EPP/Bismark cov output formats), output compression, and support for target enrichment protocols. RESULTS: Experimental results on public and simulated WGBS libraries show that our algorithmic solution is a competitive tradeoff between hash-based and BWT-based indexes, being as fast and accurate as the former, and as memory-efficient as the latter. CONCLUSIONS: The new functionalities of our bisulfite aligner and caller make it a fast and memory efficient tool, useful to analyze big datasets with little computational resources, to easily process target enrichment data, and produce statistics such as protocol efficiency and coverage as a function of the distance from target regions.

BACKGROUND: The use of short reads from High Throughput Sequencing (HTS) techniques is now commonplace in de novo assembly. Yet, obtaining contiguous assemblies from short reads is challenging, thus making scaffolding an important step in the assembly pipeline. Different algorithms have been proposed but many of them use the number of read pairs supporting a linking of two contigs as an indicator of reliability. This reasoning is intuitive, but fails to account for variation in link count due to contig features.We have also noted that published scaffolders are only evaluated on small datasets using output from only one assembler. Two issues arise from this. Firstly, some of the available tools are not well suited for complex genomes. Secondly, these evaluations provide little support for inferring a software's general performance. RESULTS: We propose a new algorithm, implemented in a tool called BESST, which can scaffold genomes of all sizes and complexities and was used to scaffold the genome of P. abies (20 Gbp). We performed a comprehensive comparison of BESST against the most popular stand-alone scaffolders on a large variety of datasets. Our results confirm that some of the popular scaffolders are not practical to run on complex datasets. Furthermore, no single stand-alone scaffolder outperforms the others on all datasets. However, BESST fares favorably to the other tested scaffolders on GAGE datasets and, moreover, outperforms the other methods when library insert size distribution is wide. CONCLUSION: We conclude from our results that information sources other than the quantity of links, as is commonly used, can provide useful information about genome structure when scaffolding.

BACKGROUND: In recent years more than 20 assemblers have been proposed to tackle the hard task of assembling NGS data. A common heuristic when assembling a genome is to use several assemblers and then select the best assembly according to some criteria. However, recent results clearly show that some assemblers lead to better statistics than others on specific regions but are outperformed on other regions or on different evaluation measures. To limit these problems we developed GAM-NGS (Genomic Assemblies Merger for Next Generation Sequencing), whose primary goal is to merge two or more assemblies in order to enhance contiguity and correctness of both. GAM-NGS does not rely on global alignment: regions of the two assemblies representing the same genomic locus (called blocks) are identified through reads' alignments and stored in a weighted graph. The merging phase is carried out with the help of this weighted graph that allows an optimal resolution of local problematic regions. RESULTS: GAM-NGS has been tested on six different datasets and compared to other assembly reconciliation tools. The availability of a reference sequence for three of them allowed us to show how GAM-NGS is a tool able to output an improved reliable set of sequences. GAM-NGS is also a very efficient tool able to merge assemblies using substantially less computational resources than comparable tools. In order to achieve such goals, GAM-NGS avoids global alignment between contigs, making its strategy unique among other assembly reconciliation tools. CONCLUSIONS: The difficulty to obtain correct and reliable assemblies using a single assembler is forcing the introduction of new algorithms able to enhance de novo assemblies. GAM-NGS is a tool able to merge two or more assemblies in order to improve contiguity and correctness. It can be used on all NGS-based assembly projects and it shows its full potential with multi-library Illumina-based projects. With more than 20 available assemblers it is hard to select the best tool. In this context we propose a tool that improves assemblies (and, as a by-product, perhaps even assemblers) by merging them and selecting the generating that is most likely to be correct.

SUMMARY: The advent of high-throughput sequencers (HTS) introduced the need of new tools in order to analyse the large amount of data that those machines are able to produce. The mandatory first step for a wide range of analyses is the alignment of the sequences against a reference genome. We present a major update to our rNA (randomized Numerical Aligner) tool. The main feature of rNA is the fact that it achieves an accuracy greater than the majority of other tools in a feasible amount of time. rNA executables and source codes are freely downloadable at http://iga-rna.sourceforge.net/. CONTACT: [email protected]; [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Next Generation Sequencing technologies are able to provide high genome coverages at a relatively low cost. However, due to limited reads' length (from 30 bp up to 200 bp), specific bioinformatics problems have become even more difficult to solve. De novo assembly with short reads, for example, is more complicated at least for two reasons: first, the overall amount of "noisy" data to cope with increased and, second, as the reads' length decreases the number of unsolvable repeats grows. Our work's aim is to go at the root of the problem by providing a pre-processing tool capable to produce (in-silico) longer and highly accurate sequences from a collection of Next Generation Sequencing reads. RESULTS: In this paper a seed-and-extend local assembler is presented. The kernel algorithm is a loop that, starting from a read used as seed, keeps extending it using heuristics whose main goal is to produce a collection of error-free and longer sequences. In particular, GapFiller carefully detects reliable overlaps and operates clustering similar reads in order to reconstruct the missing part between the two ends of the same insert. Our tool's output has been validated on 24 experiments using both simulated and real paired reads datasets. The output sequences are declared correct when the seed-mate is found. In the experiments performed, GapFiller was able to extend high percentages of the processed seeds and find their mates, with a false positives rate that turned out to be nearly negligible. CONCLUSIONS: GapFiller, starting from a sufficiently high short reads coverage, is able to produce high coverages of accurate longer sequences (from 300 bp up to 3500 bp). The procedure to perform safe extensions, together with the mate-found check, turned out to be a powerful criterion to guarantee contigs' correctness. GapFiller has further potential, as it could be applied in a number of different scenarios, including the post-processing validation of insertions/deletions detection pipelines, pre-processing routines on datasets for de novo assembly pipelines, or in any hierarchical approach designed to assemble, analyse or validate pools of sequences.

The advent of Next Generation Sequencers (NGS) has driven the necessity to design new and more sophisticated tools in order to cope with the huge amount of data produced by these novel technologies. String alignment against a genome reference is the first and most important phase in every (re)-sequencing project. Recently, distributed tools able to align large amounts of sequences using clusters or clouds of computers, have been put forward. The aim of this work is to propose a new tool named mrNA (the MPI version of the original rNA program) able to align NGS data using a cluster of computers. mrNA was designed to tackle the main computational bottleneck of all classical parallel implementation of aligners: references longer than 4 Gbp. mrNA, together with rNA, are open source programs downloadable at http://iga-rna.sourceforge.net/.

Next Generation Sequencing has totally changed genomics: we are able to produce huge amounts of data at an incredible low cost if compared to Sanger sequencing. Despite this some old problems have become even more difficult, de-novo assembly being on top of this list. In this paper we propose a novel method that aims at improving de-novo assembly in presence of a closely related reference. The idea is to combine de-novo assembly and reference guided assembly in order to obtain an enhanced assembly.