Protein molecules differ in flexibility across their three-dimensional configurations. The phi ($\varphi$) and psi ($\psi$) torsion angles primarily define the protein backbone. The flexibility of proteins is related to the fluctuation of the torsion angle. The fluctuation of torsion angles is caused by the differences in backbone torsion angles between various NMR models. The angle fluctuations in the cartesian coordinate space are utilized to define the structural flexibility of proteins and help predict protein function and structure when the torsion angles are employed as constraints. This research attempts to develop a machine-learning method for directly predicting fluctuations in the torsion angle of protein sequences. We collect a number of helpful characteristics of proteins, including disorder probability, position-specific scoring matrix profiles, secondary structure probabilities, monograms, bigrams, position-specific estimated energy, half-sphere exposures. Similarly, we explore well-known machine learning algorithms and present an optimized Light Gradient Boosting Machine Regressor (LightGBM) method, named TAFPred, to predict torsion angle variations using the selected features. The proposed method achieves ten-fold cross-validated correlation coefficients of0.746 and 0.737, as well as mean absolute errors of 0.114 and 0.123, for the angle fluctuation of $\varphi$ and $\psi$, respectively, and an improvement of 6.59% in MAE, 24.50% in PCC in the phi angle, and 6.09% in MAE, 21.84% in PCC in the psi angle, compared to the state-of-the-art method proposed by Zhang et al.

Marine mammal vocalizations provide a reliable method of identifying most species. These vocalizations may be effectively captured by the passive acoustic monitoring of underwater acoustic signals, but this approach generates prohibitively large amounts of data, making a machine learning approach desirable. In this study, we demonstrate that residual learning networks may be highly effective at classifying vocalizations by species and investigate the utility of multi-channel spectrograms in training these networks. We find that classification performance can be improved by effectively preprocessing the acoustic data. A series of tests are used to assess optimal parameters for noise-removal, spectrographic window functions, preprocessing augmentations, and multi-channel spectrogram generation. While we find that our networks do learn additional information from multi-channel spectrograms, we demonstrate that single-channel spectrograms still offer superior classification performance in 32-class classification tasks.

Levee systems require continuous monitoring and maintenance because of the risk of failure causing catastrophic flooding. This study explores several image segmentation methods based on fully convolutional neural networks to automatically detect cracks from images at the pixel level in the levee systems. Cracks are the leading cause of the failure of these flood water control structures. This research aims to use a levee crack dataset to train and evaluate the seven latest deep learning-based image segmentation models. With the training setup, the MultiResUnet model achieved the highest mean Intersection over Union (mIoU) of 87.777% on 10-Fold Cross-Validation and 87.766% on hold-out test data compared to other models.

MOTIVATION: Transposable Elements (TEs) or jumping genes are DNA sequences that have an intrinsic capability to move within a host genome from one genomic location to another. Studies show that the presence of a TE within or adjacent to a functional gene may alter its expression. TEs can also cause an increase in the rate of mutation and can even mediate duplications and large insertions and deletions in the genome, promoting gross genetic rearrangements. The proper classification of identified jumping genes is important for analyzing their genetic and evolutionary effects. An effective classifier, which can explain the role of TEs in germline and somatic evolution more accurately, is needed. In this study, we examine the performance of a variety of machine learning (ML) techniques and propose a robust method, ClassifyTE, for the hierarchical classification of TEs with high accuracy, using a stacking-based ML method. RESULTS: We propose a stacking-based approach for the hierarchical classification of TEs. When trained on three different benchmark datasets, our proposed system achieved 4%, 10.68% and 10.13% average percentage improvement (using the hF measure) compared to several state-of-the-art methods. We developed an end-to-end automated hierarchical classification tool based on the proposed approach, ClassifyTE, to classify TEs up to the super-family level. We further evaluated our method on a new TE library generated by a homology-based classification method and found relatively high concordance at higher taxonomic levels. Thus, ClassifyTE paves the way for a more accurate analysis of the role of TEs. AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/manisa/ClassifyTE. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

We present a novel approach for the prediction of Rogue Waves in oceans using Machine Learning methods. Since the ocean is composed of many wave systems, the change from a bimodal or multimodal directional distribution to unimodal one is taken as the warning criteria. Further, we explore various features that help in predicting Rogue Waves. The analysis of the results shows that the Spectral features are significant in predicting Rogue Waves. Finally, we propose a Random Forest Classifier based algorithm to predict Rogue Waves in oceanic conditions. For a range of windows, the proposed algorithm has accuracies between 89.57% and 91.81%, and the balanced accuracies between 79.41% and 89.03%. Moreover, we have also introduced the publicly available buoy dataset, which can serve as a benchmark dataset.

This paper describes and compares different approaches for detecting cracks in the concrete toe, other general areas of levees and dams using satellite or drone images. The dataset was sourced from real drone flight data and manually collected and annotated as needed. We compare old and modern algorithms alike to determine which ones perform best in this case. We also explain the reasonings for a particularly interesting case of the viola jones algorithm, where our calculated accuracy is 100%. We study stacking (85% accuracy), and the latest deep learning techniques (90.90% accuracy) as well. This research hopes to help the U.S. Army Engineers Corps integrate the model into drones to better monitor the levee areas prone to disaster.

Understanding and predicting weather behavior is vital for informing pilots about changing flight conditions. This paper presents a new approach towards forecasting one component of weather information, wind speed, from data captured by airplanes in flight. We compare two datasets for prediction suitability, and a collinearity analysis between these datasets reveals a better model performance with smaller test error with one of them. We then apply machine learning and a genetic algorithm to process this data further and arrive at a competitive error rate. Finally, we create an offline software for wind prediction using the best performing classifier.

Motivation: Identification of DNA-binding proteins from only sequence information is one of the most challenging problems in the field of genome annotation. DNA-binding proteins play an important role in various biological processes such as DNA replication, repair, transcription and splicing. Existing experimental techniques for identifying DNA-binding proteins are time-consuming and expensive. Thus, prediction of DNA-binding proteins from sequences alone using computational methods can be useful to quickly annotate and guide the experimental process. Most of the methods developed for predicting DNA-binding proteins use the information from the evolutionary profile, called the position-specific scoring matrix (PSSM) profile, alone and the accuracies of such methods have been limited. Here, we propose a method, called StackDPPred, which utilizes features extracted from PSSM and residue specific contact-energy to help train a stacking based machine learning method for the effective prediction of DNA-binding proteins. Results: Based on benchmark sequences of 1063 (518 DNA-binding and 545 non DNA-binding) proteins and using jackknife validation, StackDPPred achieved an ACC of 89.96%, MCC of 0.799 and AUC of 94.50%. This outcome outperforms several state-of-the-art approaches. Furthermore, when tested on recently designed two independent test datasets, StackDPPred outperforms existing approaches consistently. The proposed StackDPPred can be used for effective prediction of DNA-binding proteins from sequence alone. Availability and implementation: Online server is at http://bmll.cs.uno.edu/add and code-data is at http://cs.uno.edu/∼tamjid/Software/StackDPPred/code_data.zip. Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Machine learning plays a substantial role in bioscience owing to the explosive growth in sequence data and the challenging application of computational methods. Peptide-recognition domains (PRDs) are critical as they promote coupled-binding with short peptide-motifs of functional importance through transient interactions. It is challenging to build a reliable predictor of peptide-binding residue in proteins with diverse types of PRDs from protein sequence alone. On the other hand, it is vital to cope up with the sequencing speed and to broaden the scope of study. Results: In this paper, we propose a machine-learning-based tool, named PBRpredict, to predict residues in peptide-binding domains from protein sequence alone. To develop a generic predictor, we train the models on peptide-binding residues of diverse types of domains. As inputs to the models, we use a high-dimensional feature set of chemical, structural and evolutionary information extracted from protein sequence. We carefully investigate six different state-of-the-art classification algorithms for this application. Finally, we use the stacked generalization approach to non-linearly combine a set of complementary base-level learners using a meta-level learner which outperformed the winner-takes-all approach. The proposed predictor is found competitive based on statistical evaluation. Availability and implementation: PBRpredict-Suite software: http://cs.uno.edu/~tamjid/Software/PBRpredict/pbrpredict-suite.zip. Supplementary information: Supplementary data are available at Bioinformatics online.

In-vitro methods for protein structure determination are time-consuming, cost-intensive, and failure-prone. Because of these expenses, alternative computer-based predictive methods have emerged. Predicting a protein's 3-D structure from only its amino acid sequence-also known as ab initio protein structure prediction (PSP)-is computationally demanding because the search space is astronomically large and energy models are extremely complex. Some successes have been achieved in predictive methods but these are limited to small sized proteins (around 100 amino acids); thus, developing efficient algorithms, reducing the search space, and designing effective search guidance heuristics are necessary to study large sized proteins. An on-lattice model can be a better ground for rapidly developing and measuring the performance of a new algorithm, and hence we consider this model for larger proteins (>150 amino acids) to enhance the genetic algorithms (GAs) framework. In this paper, we formulate PSP as a combinatorial optimization problem that uses 3-D face-centered-cubic lattice coordinates to reduce the search space and hydrophobic-polar energy model to guide the search. The whole optimization process is controlled by an enhanced GA framework with four enhanced features: 1) an exhaustive generation approach to diversify the search; 2) a novel hydrophobic core-directed macro-mutation operator to intensify the search; 3) a per-generation duplication elimination strategy to prevent early convergence; and 4) a random-walk technique to recover from stagnation. On a set of standard benchmark proteins, our algorithm significantly outperforms state-of-the-art algorithms. We also experimentally show that our algorithm is robust enough to produce very similar results regardless of different parameter settings.

No single algorithm suits the best for the protein structure prediction problem. Therefore, researchers have tried hybrid techniques to mix the power of different strategies to gain improvements. In this paper, we present a hybrid search framework that embeds a tabu-based local search within a population based genetic algorithm. We applied our hybrid algorithm on simplified protein structure prediction problem. We use a low-resolution ab initio search method with the hydrophobic-polar energy model and face-centred-cubic lattice. Within the genetic algorithm, we apply local search in two different situations: i) only once at the beginning and ii) every time at search stagnation. At the beginning, we apply local search to improve the randomly generated individuals and use them as an initial population for the genetic algorithm. Later, we apply local search after applying a random-walk at situations where the genetic algorithm gets stuck. In both cases, the use of local search is to improve the randomised solutions quickly. We experimentally show that our hybrid approach outperforms the state-of-the-art approaches.

BACKGROUND: Protein structure prediction is an important but unsolved problem in biological science. Predicted structures vary much with energy functions and structure-mapping spaces. In our simplified ab initio protein structure prediction methods, we use hydrophobic-polar (HP) energy model for structure evaluation, and 3-dimensional face-centred-cubic lattice for structure mapping. For HP energy model, developing a compact hydrophobic-core (H-core) is essential for the progress of the search. The H-core helps find a stable structure with the lowest possible free energy. RESULTS: In order to build H-cores, we present a new Spiral Search algorithm based on tabu-guided local search. Our algorithm uses a novel H-core directed guidance heuristic that squeezes the structure around a dynamic hydrophobic-core centre. We applied random walks to break premature H-cores and thus to avoid early convergence. We also used a novel relay-restart technique to handle stagnation. CONCLUSIONS: We have tested our algorithms on a set of benchmark protein sequences. The experimental results show that our spiral search algorithm outperforms the state-of-the-art local search algorithms for simplified protein structure prediction. We also experimentally show the effectiveness of the relay-restart.

To gain a better understanding of how proteins function a process known as protein structure prediction (PSP) is carried out. However, experimental PSP methods, such as X-ray crystallography and Nuclear Magnetic Resonance (NMR), can be time-consuming and inaccurate. This has given rise to numerous computational PSP approaches to try and elicit a protein's three-dimensional conformation. A popular PSP search strategy is Genetic Algorithms (GA). GAs allow for a generic search approach, which can provide a generic improvement to alleviate the need to redefine the search strategies for separate sequences. Though GA's working principles are remarkable, a serious problem that is inherent in the GA search process is the growth of twins or identical chromosomes. Therefore, enhanced twin removal strategies are crucial for any GA search solving hard-optimisation problems like PSP. In this paper we explain our high-resolution GA feature-based resampling PSP approach and propose a twin removal strategy to further enhance its prediction accuracy. This includes investigating the optimal chromosome correlation factor (CCF) for our approach and defining a pre-built structure library for twin removal. We have also compared our GA approach with the popular Monte Carlo (MC) method for PSP. Our results indicate that out of all the CCF values we tested a CCF value of 0.8 provided the best level of diversity within our GA population. It also generated, on average, more native-like structures than any of the other CCF values, and clearly demonstrated that twin removal is needed in PSP when using GAs to obtain more accurate results.

This paper presents the impact of twins and the measures for their removal from the population of genetic algorithm (GA) when applied to effective conformational searching. It is conclusively shown that a twin removal strategy for a GA provides considerably enhanced performance when investigating solutions to complex ab initio protein structure prediction (PSP) problems in low-resolution model. Without twin removal, GA crossover and mutation operations can become ineffectual as generations lose their ability to produce significant differences, which can lead to the solution stalling. The paper relaxes the definition of chromosomal twins in the removal strategy to not only encompass identical, but also highly correlated chromosomes within the GA population, with empirical results consistently exhibiting significant improvements solving PSP problems.

Proteins carry out the majority of functionality on a cellular level. Computational protein structure prediction (PSP) methods have been introduced to speed up the PSP process due to manual methods, like nuclear magnetic resonance (NMR) and x-ray crystallography (XC) taking numerous months even years to produce a predicted structure for a target protein. A lot of work in this area is focused on the type of search strategy to employ. Two popular methods in the literature are: Monte Carlo based algorithms and Genetic Algorithms. Genetic Algorithms (GA) have proven to be quite useful in the PSP field, as they allow for a generic search approach, which alleviates the need to redefine the search strategies for separate sequences. They also lend themselves well to feature-based resampling techniques. Feature-based resampling works by taking previously computed local minima and combining features from them to create new structures that are more uniformly low in free energy. In this work we present a feature-based resampling genetic algorithm to refine structures that are outputted by PSP software. Our results indicate that our approach performs well, and produced an average 9.5% root mean square deviation (RMSD) improvement and a 17.36% template modeling score (TM-Score) improvement.

In most of the successful real protein structure prediction (PSP) problem, lattice models have been essentially utilized to have the folding backbone sampling at the top of the hierarchical approach. A three dimensional face-centred-cube (FCC), with the provision for providing the most compact core, can map closest to the folded protein in reality. Hence, our successful hybrid genetic algorithms (HGA) proposed earlier for a square and cube lattice model is being extended in this paper for a 3D FCC model. Furthermore, twins (conformations having similarity with each other), in GA population have also been considered for removal from the search space for improving the effectiveness of GA The HGA combined with the twin removal (TR) strategy showed best performance when compared with the simple GA (SGA), SGA with TR, and HGA only versions. Experiments were carried out on the publicly available benchmark HP sequences and results are expressed based on the fitness of the corresponding applied lattice model, which will help any future novel approach to be compared.

In this paper, a Guided Genetic Algorithm (GGA) has been presented for protein folding prediction (PFP) using 3D Hydrophobic-Hydrophilic (HP) model. Effective strategies have been formulated utilizing the core formation of the globular protein, which provides the guideline for the Genetic Algorithm (GA) while predicting protein folding. Building blocks containing Hydrophobic (H) -Hydrophilic (P or Polar) covalent bond are utilized such a way that it helps form a core that maximizes the fitness. A series of operators are developed including Diagonal Move and Tilt Move to assist in implementing the building blocks in three-dimensional space. The GGA outperformed Unger's GA in 3D HP model. The overall strategy incorporates a swing function that provides a mechanism to enable the GGA to test more potential solutions and also prevent it from developing a schema that may cause it to become trapped in local minima. Further, it helps the guidelines remain non-rigid. GGA provides improved and robust performance for PFP.

Traditional encodings for hydrophobic(H)-hydrophilic(P) model or HP lattice models is isomorphic, which adds unwanted variations for the same solution, thereby slowing convergence. In this paper a novel non-isomorphic encoding scheme is presented for HP lattice model, which constrains the search space. In addition, similarity comparisons are made easier and more consistent and it will be shown that non-deterministic search approach such as genetic algorithm (GA) converges faster when non-isomorphic encoding is employed

This paper presents a novel guided genetic algorithm (GGA) for protein folding prediction (PFP) in 2D hydrophobic-hydrophilic (HP) by exploring the protein core formation concept. A proof of the shape for an optimal core is provided and a set of highly probable sub-conformations are defined which help to establish the guidelines to form the core boundary. A series of new operators including diagonal move and tilt move are defined to assist in implementing the guidelines. The underlying reasons for the failure in the folding prediction of relatively long sequences using Unger's genetic algorithm (GA) in 2D HP model are analysed and the new GGA is shown to overcome these limitations. The overall strategy incorporates a swing function that provides a mechanism to enable the GGA to test more potential solutions and also prevent it from developing a schema that may cause it to become trapped in local minima. While the guidelines do not force particular conformations, the result is a number of conformations for particular putative ground energy and superior prediction accuracy, endorsing the improved performance compared with other well established nondeterministic search approaches

The protein folding problem is a minimization problem in which the energy function is often regarded as the fitness function. There are several models for protein folding prediction including the hydrophobic-hydrophilic (HP) model. Though this model is an elementary one, it is widely used as a test-bed for faster execution of new algorithms. Fitness computation is one of the major computational parts of the HP model. This paper proposes an efficient search (ES) approach for computing the fitness value requiring only O(n) complexity in contrast to the full search (FS) approach that requires O(n/sup 2/) complexity. The efficiency of the proposed ES approach results due to its utilization of some inherent properties of the HP model. The ES approach represents residues in a Cartesian coordinate framework and then uses relative distance and coordinate polarity to reduce complexity.

Fitness computation after each crossover or mutation operation in genetic algorithm (GA) requires computational time that increases with the increasing length of the chromosome. In this paper, an efficient GA is proposed for protein folding prediction based on the hydrophobic-hydrophilic (HP) model. The partial fitness of the parent computed from one end of sequence till crossover or mutation point is utilized for the computation of the fitness of the child. The calculated value of the partial fitness is stored with the corresponding chromosome. Although the approach requires additional memory for each hydrophobic residue of each chromosome, the computation time is reduced significantly, which is more important than the memory overhead.