Biology subjects

Pinho, A. J.

Publications and source records attributed to Pinho, A. J..

7 recordsLinked to original sources

bioRxiv2025

Automated Annotation of Plant Gene Regions Using Supervised Machine Learning

Rapid sequencing now enables analysis across many species, including plants, but repeat-rich, redundant plant genomes still hinder precise annotation. Precise gene calls are vital for understanding biology, comparative genomics, and discovering traits such as improved nutrition and disease resistance. We present GeAnno, a supervised machine learning method for plant gene detection. GeAnno uses an XGBoost classifier trained on curated plant annotations and a sliding-window scheme while capturing redundancy, base composition, and start/stop-codon spacing, among others. At inference, windows are scored as genic or intergenic and lightly smoothed. The output is a standard GFF3 with strand-specific genic regions. Benchmarking against ab initio predictors under matched training on 11 cassava (Manihot esculenta) genomes and cross-species evaluation, GeAnno achieved higher nucleotide-level precision and F1-score. For example, on cassava it reached 77.13% precision and 72.90% F1-score. Moreover, performance transfers to divergent species and parameters (window, step, smoothing, thresholds) are tunable. By improving accuracy and portability on complex plant genomes, GeAnno supports downstream functional studies and breeding, advancing food security and ecological sustainability.

bioinformatics↗

bioRxiv2025

Metagenomic Classification of Ancient Viruses

AO_SCPLOWBSTRACTC_SCPLOWAncient DNA (aDNA) sequences present unique challenges for taxonomic classification due to extreme fragmentation (reads 20-100 bp), end-biased cytosine deamination, and high contamination rates. Conventional metagenomic classifiers based on exact k-mer matching or alignment lose discriminative power on such short and damaged reads, limiting the analysis of paleogenomic samples. We present FALCON2, a compression-based metagenomic classifier that leverages position-aware finite-context models to maintain high accuracy on degraded viral ancient viruses. FALCON2 consolidates the capabilities of its predecessor, FALCON-meta, into a unified executable with enhanced features including model persistence, direct processing of compressed inputs, multiple file handling, and optional pre-filtering methodologies for contaminated samples. Under controlled benchmarking with database, taxonomy, and thread parity on simulated viral datasets, FALCON2 achieved an Area Under the Curve of Receiver Operating Characteristic (AUC-ROC) of 0.999, an Area Under Precision-Recall Curve (AUPRC) of 0.968, and an F1-score of 0.918, substantially outperforming Centrifuge (AUPRC = 0.625), Kraken2 (AUPRC = 0.184), and CLARK-S (AUPRC = 0.013) on pooled micro-averaged metrics. FALCON2s advantage is most pronounced on ultra-short reads (20-40 bp), where exact k-mers become sparse. FALCON2 pre-filtering at threshold 0.7 improved precision by 10 percentage points with negligible recall loss. FALCON2 runs on systems with 4-8 GB RAM for typical analyses. FALCON2 is freely available at https://github.com/cobilab/FALCON2 under GPL v3 license.

bioinformatics↗

bioRxiv2025

Optimizing Genomic Data Compression with Genetic Algorithms

AO_SCPLOWBSTRACTC_SCPLOWThe rapid growth of genomic data, especially in clinical settings, has highlighted the need for efficient lossless compression. Preserving full sequence information is critical, as even small data losses can affect analyses or diagnoses. With increasing dataset size and complexity, optimized compression is vital for storage, transmission, and processing. Fixed-parameter compression tools often fall short across diverse genomes, demonstrating the need for adaptive tuning strategies tailored to each dataset. In this paper, we investigate the use of metaheuristic search algorithms, particularly Genetic Algorithms (GAs), to optimize parameter configurations for JARVIS3, a genomic sequence compression tool. Due to JARVIS3s large and complex parameter space, exhaustive search is infeasible. To tackle this, we developed OptimJV3, a modular framework that integrates multiple GA variants, including a multi-objective approach that also considers computational time. We evaluated OptimJV3 on various genomic datasets (FASTA format), namely Human Chromosome Y (CY), Cassava, and the full Human Genome (HG), and observed notable compression gains. For large sequences like Cassava and HG, we applied sampling-based optimization: optimal parameters were first identified from smaller segments and then used for full sequence compression. These small segments proved effective for guiding parameter tuning. Further improvements were achieved by increasing the number of GA generations, especially for HG. For example, using a 10 MB sample and 500 generations, OptimJV3 reached a compression rate of 1.431 bits per base on HG - among the best results reported so far.

bioinformatics↗

bioRxiv2023

AlcoR: alignment-free simulation, mapping, and visualization of low-complexity regions in biological data

Low-complexity data analysis is the area that addresses the search and quantification of regions in sequences of elements that contain low-complexity or repetitive elements. For example, these can be tandem repeats, inverted repeats, homopolymer tails, GC biased regions, similar genes, and hairpins, among many others. Identifying these regions is crucial because of their association with regulatory and structural characteristics. Moreover, their identification provides positional and quantity information where standard assembly methodologies face significant difficulties because of substantial higher depth coverage (mountains), ambiguous read mapping, or where sequencing or reconstruction defects may occur. However, the capability to distinguish low-complexity regions (LCRs) in genomic and proteomic sequences is a challenge that depends on the models ability to find them automatically. Low-complexity patterns can be implicit through specific or combined sources, such as algorithmic or probabilistic, and recurring to different spatial distances, namely local, medium, or distant associations.This paper addresses the challenge of automatically modeling and distinguishing LCRs, providing a new method and tool (AlcoR) for efficient and accurate segmentation and visualization of these regions in genomic and proteomic sequences. The method enables the use of models with different memories, providing the ability to distinguish local from distant low-complexity patterns. The method is reference- and alignment-free, providing additional methodologies for testing, including a highly-flexible simulation method for generating biological sequences (DNA or protein) with different complexity levels, sequence masking, and a visualization tool for automatic computation of the LCR maps into an ideogram style. We provide illustrative demonstrations using synthetic, nearly synthetic, and natural sequences showing the high efficiency and accuracy of AlcoR. As large-scale results, we use AlcoR to unprecedentedly provide a whole-chromosome low-complexity map of a recent complete human genome and the haplotype-resolved chromosome pairs of a heterozygous diploid African cassava cultivar.The AlcoR method provides the ability of fast sequence characterization through data complexity analysis, ideally for scenarios entangling the presence of new or unknown sequences. AlcoR is implemented in C language using multi-threading to increase the computational speed, is flexible for multiple applications, and does not contain external dependencies. The tool accepts any sequence in FASTA format. The source code is freely provided at https://github.com/cobilab/alcor.

bioinformatics↗

bioRxiv2020

GTO: a toolkit to unify pipelines in genomic and proteomic research

SummaryNext-generation sequencing triggered the production of a massive volume of publicly available data and the development of new specialised tools. These tools are dispersed over different frameworks, making the management and analyses of the data a challenging task. Additionally, new targeted tools are needed, given the dynamics and specificities of the field. We present GTO, a comprehensive toolkit designed to unify pipelines in genomic and proteomic research, which combines specialised tools for analysis, simulation, compression, development, visualisation, and transformation of the data. This toolkit combines novel tools with a modular architecture, being an excellent platform for experimental scientists, as well as a useful resource for teaching bioinformatics inquiry to students in life sciences. Availability and implementationGTO is implemented in C language and it is available, under the MIT license, at http://bioinformatics.ua.pt/gto. Contactpratas@ua.pt Supplementary informationSupplementary data are available at publishers Web site.

bioinformatics↗

bioRxiv2019

Smash++: an alignment-free and memory-efficient tool to find genomic rearrangements

BackgroundThe development of high-throughput sequencing technologies and, as its result, the production of huge volumes of genomic data, has accelerated biological and medical research and discovery. Study on genomic rearrangements is crucial due to their role in chromosomal evolution, genetic disorders and cancer; ResultsWe present Smash++, an alignment-free and memory-efficient tool to find and visualize small- and large-scale genomic rearrangements between two DNA sequences. This computational solution extracts information contents of the two sequences, exploiting a data compression technique, in order for finding rearrangements. We also present Smash++ visualizer, a tool that allows the visualization of the detected rearrangements along with their self- and relative complexity, by generating an SVG (Scalable Vector Graphics) image; ConclusionsTested on several synthetic and real DNA sequences from bacteria, fungi, Aves and mammalia, the proposed tool was able to accurately find genomic rearrangements. The detected regions complied with previous studies which took alignment-based approaches or performed FISH (Fluorescence in situ hybridization) analysis. The maximum peak memory usage among all experiments was ~1 GB, which makes Smash++ feasible to run on present-day standard computers.

bioinformatics↗

bioRxiv2018

FALCON: a method to infer metagenomic composition of ancient DNA

The general approaches to detect and quantify metagenomic sample composition are based on the alignment of the reads, according to an existing database containing reference microbial sequences. However, without proper parameterization, these methods are not suitable for ancient DNA. Quantifying somewhat dissimilar sequences by alignment methods is problematic, due to the need of fine-tuned thresholds, considering relaxed edit distances and the consequent increase of computational cost. Additionally, the choice of the thresholds poses the problem of how to quantify similarity without producing overestimated measures. We propose FALCON-meta, a compression-based method to infer metagenomic composition of next-generation sequencing samples. This unsupervised alignment-free method runs efficiently on FASTQ samples. FALCON-meta quickly learns how to give importance to the models that cooperate to predict similarity, incorporating parallelism and flexibility for multiple hardware characteristics. It shows substantial identification capabilities in ancient DNA without overestimation. In one of the examples, we found and authenticated an ancient Pseudomonas bacteria in a Mammoth mitogenome.\n\nFALCON-meta can be accessed at https://github.com/pratas/falcon.

bioinformatics↗