Search for: All records

Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity, and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries.

Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns.

All deep sequencing datasets and code to perform the analyses presented within are available via https://github.com/Tessier-Lab-UMich/PSERM_paper.

Timetrees depict evolutionary relationships between species and the geological times of their divergence. Hundreds of research articles containing timetrees are published in scientific journals every year. The TimeTree (TT) project has been manually locating, curating and synthesizing timetrees from these articles for almost two decades into a TimeTree of Life, delivered through a unique, user-friendly web interface (timetree.org). The manual process of finding articles containing timetrees is becoming increasingly expensive and time-consuming. So, we have explored the effectiveness of text-mining approaches and developed optimizations to find research articles containing timetrees automatically.

We have developed an optimized machine learning system to determine if a research article contains an evolutionary timetree appropriate for inclusion in the TT resource. We found that BERT classification fine-tuned on whole-text articles achieved an F1 score of 0.67, which we increased to 0.88 by text-mining article excerpts surrounding the mentioning of figures. The new method is implemented in the TimeTreeFinder (TTF) tool, which automatically processes millions of articles to discover timetree-containing articles. We estimate that the TTF tool would produce twice as many timetree-containing articles as those discovered manually, whose inclusion in the TT database would potentially double the knowledge accessible to a wider community. Manual inspection showed that the precision on out-of-distribution recently published articles is 87%. This automation will speed up the collection and curation of timetrees with much lower human and time costs.

https://github.com/marija-stanojevic/time-tree-classification.

Supplementary data are available at Bioinformatics online.

Over the past decade, short-read sequence alignment has become a mature technology. Optimized algorithms, careful software engineering and high-speed hardware have contributed to greatly increased throughput and accuracy. With these improvements, many opportunities for performance optimization have emerged. In this review, we examine three general-purpose short-read alignment tools—BWA-MEM, Bowtie 2 and Arioc—with a focus on performance optimization. We analyze the performance-related behavior of the algorithms and heuristics each tool implements, with the goal of arriving at practical methods of improving processing speed and accuracy. We indicate where an aligner's default behavior may result in suboptimal performance, explore the effects of computational constraints such as end-to-end mapping and alignment scoring threshold, and discuss sources of imprecision in the computation of alignment scores and mapping quality. With this perspective, we describe an approach to tuning short-read aligner performance to meet specific data-analysis and throughput requirements while avoiding potential inaccuracies in subsequent analysis of alignment results. Finally, we illustrate how this approach avoids easily overlooked pitfalls and leads to verifiable improvements in alignment speed and accuracy.

richard.wilton@jhu.edu

Appendices referenced in this article are available at Bioinformatics online.

Properties of molecules are indicative of their functions and thus are useful in many applications. With the advances of deep-learning methods, computational approaches for predicting molecular properties are gaining increasing momentum. However, there lacks customized and advanced methods and comprehensive tools for this task currently.

Here, we develop a suite of comprehensive machine-learning methods and tools spanning different computational models, molecular representations and loss functions for molecular property prediction and drug discovery. Specifically, we represent molecules as both graphs and sequences. Built on these representations, we develop novel deep models for learning from molecular graphs and sequences. In order to learn effectively from highly imbalanced datasets, we develop advanced loss functions that optimize areas under precision–recall curves (PRCs) and receiver operating characteristic (ROC) curves. Altogether, our work not only serves as a comprehensive tool, but also contributes toward developing novel and advanced graph and sequence-learning methodologies. Results on both online and offline antibiotics discovery and molecular property prediction tasks show that our methods achieve consistent improvements over prior methods. In particular, our methods achieve #1 ranking in terms of both ROC-AUC (area under curve) and PRC-AUC on the AI Cures open challenge for drug discovery related to COVID-19.

Our source code is released as part of the MoleculeX library (https://github.com/divelab/MoleculeX) under AdvProp.

Supplementary data are available at Bioinformatics online.

Cell shape provides both geometry for, and a reflection of, cell function. Numerous methods for describing and modeling cell shape have been described, but previous evaluation of these methods in terms of the accuracy of generative models has been limited.

Here we compare traditional methods and deep autoencoders to build generative models for cell shapes in terms of the accuracy with which shapes can be reconstructed from models. We evaluated the methods on different collections of 2D and 3D cell images, and found that none of the methods gave accurate reconstructions using low dimensional encodings. As expected, much higher accuracies were observed using high dimensional encodings, with outline-based methods significantly outperforming image-based autoencoders. The latter tended to encode all cells as having smooth shapes, even for high dimensions. For complex 3D cell shapes, we developed a significant improvement of a method based on the spherical harmonic transform that performs significantly better than other methods. We obtained similar results for the joint modeling of cell and nuclear shape. Finally, we evaluated the modeling of shape dynamics by interpolation in the shape space. We found that our modified method provided lower deformation energies along linear interpolation paths than other methods. This allows practical shape evolution in high dimensional shape spaces. We conclude that our improved spherical harmonic based methods are preferable for cell and nuclear shape modeling, providing better representations, higher computational efficiency and requiring fewer training images than deep learning methods.

All software and data is available at http://murphylab.cbd.cmu.edu/software.

Supplementary data are available at Bioinformatics online.

Due to the nature of experimental annotation, most protein function prediction methods operate at the protein-level, where functions are assigned to full-length proteins based on overall similarities. However, most proteins function by interacting with other proteins or molecules, and many functional associations should be limited to specific regions rather than the entire protein length. Most domain-centric function prediction methods depend on accurate domain family assignments to infer relationships between domains and functions, with regions that are unassigned to a known domain-family left out of functional evaluation. Given the abundance of residue-level annotations currently available, we present a function prediction methodology that automatically infers function labels of specific protein regions using protein-level annotations and multiple types of region-specific features.

We apply this method to local features obtained from InterPro, UniProtKB and amino acid sequences and show that this method improves both the accuracy and region-specificity of protein function transfer and prediction. We compare region-level predictive performance of our method against that of a whole-protein baseline method using proteins with structurally verified binding sites and also compare protein-level temporal holdout predictive performances to expand the variety and specificity of GO terms we could evaluate. Our results can also serve as a starting point to categorize GO terms into region-specific and whole-protein terms and select prediction methods for different classes of GO terms.

The code and features are freely available at: https://github.com/ek1203/rsfp.

Supplementary data are available at Bioinformatics online.

Adverse drug reactions (ADRs) are one of the main causes of death and a major financial burden on the world’s economy. Due to the limitations of the animal model, computational prediction of serious and rare ADRs is invaluable. However, current state-of-the-art computational methods do not yield significantly better predictions of rare ADRs than random guessing.

We present a novel method, based on the theory of ‘compressed sensing’ (CS), which can accurately predict serious side-effects of candidate and market drugs. Not only is our method able to infer new chemical-ADR associations using existing noisy, biased and incomplete databases, but our data also demonstrate that the accuracy of CS in predicting a serious ADR for a candidate drug increases with increasing knowledge of other ADRs associated with the drug. In practice, this means that as the candidate drug moves up the different stages of clinical trials, the prediction accuracy of our method will increase accordingly.

The program is available at https://github.com/poleksic/side-effects.

Supplementary data are available at Bioinformatics online.

The prevalence of high-throughput experimental methods has resulted in an abundance of large-scale molecular and functional interaction networks. The connectivity of these networks provides a rich source of information for inferring functional annotations for genes and proteins. An important challenge has been to develop methods for combining these heterogeneous networks to extract useful protein feature representations for function prediction. Most of the existing approaches for network integration use shallow models that encounter difficulty in capturing complex and highly non-linear network structures. Thus, we propose deepNF, a network fusion method based on Multimodal Deep Autoencoders to extract high-level features of proteins from multiple heterogeneous interaction networks.

We apply this method to combine STRING networks to construct a common low-dimensional representation containing high-level protein features. We use separate layers for different network types in the early stages of the multimodal autoencoder, later connecting all the layers into a single bottleneck layer from which we extract features to predict protein function. We compare the cross-validation and temporal holdout predictive performance of our method with state-of-the-art methods, including the recently proposed method Mashup. Our results show that our method outperforms previous methods for both human and yeast STRING networks. We also show substantial improvement in the performance of our method in predicting gene ontology terms of varying type and specificity.

deepNF is freely available at: https://github.com/VGligorijevic/deepNF.

Supplementary data are available at Bioinformatics online.

Accurately predicting drug–target interactions (DTIs) in silico can guide the drug discovery process and thus facilitate drug development. Computational approaches for DTI prediction that adopt the systems biology perspective generally exploit the rationale that the properties of drugs and targets can be characterized by their functional roles in biological networks.

Inspired by recent advance of information passing and aggregation techniques that generalize the convolution neural networks to mine large-scale graph data and greatly improve the performance of many network-related prediction tasks, we develop a new nonlinear end-to-end learning model, called NeoDTI, that integrates diverse information from heterogeneous network data and automatically learns topology-preserving representations of drugs and targets to facilitate DTI prediction. The substantial prediction performance improvement over other state-of-the-art DTI prediction methods as well as several novel predicted DTIs with evidence supports from previous studies have demonstrated the superior predictive power of NeoDTI. In addition, NeoDTI is robust against a wide range of choices of hyperparameters and is ready to integrate more drug and target related information (e.g. compound–protein binding affinity data). All these results suggest that NeoDTI can offer a powerful and robust tool for drug development and drug repositioning.

The source code and data used in NeoDTI are available at: https://github.com/FangpingWan/NeoDTI.

Supplementary data are available at Bioinformatics online.

Genome-scale metabolic network models and constraint-based modeling techniques have become important tools for analyzing cellular metabolism. Thermodynamically infeasible cycles (TICs) causing unbounded metabolic flux ranges are often encountered. TICs satisfy the mass balance and directionality constraints but violate the second law of thermodynamics. Current practices involve implementing additional constraints to ensure not only optimal but also loopless flux distributions. However, the mixed integer linear programming problems required to solve become computationally intractable for genome-scale metabolic models.

We aimed to identify the fewest needed constraints sufficient for optimality under the loopless requirement. We found that loopless constraints are required only for the reactions that share elementary flux modes representing TICs with reactions that are part of the objective function. We put forth the concept of localized loopless constraints (LLCs) to enforce this minimal required set of loopless constraints. By combining with a novel procedure for minimal null-space calculation, the computational time for loopless flux variability analysis (ll-FVA) is reduced by a factor of 10–150 compared to the original loopless constraints and by 4–20 times compared to the current fastest method Fast-SNP with the percent improvement increasing with model size. Importantly, LLCs offer a scalable strategy for loopless flux calculations for multi-compartment/multi-organism models of large sizes, for example, shortening the CPU time for ll-FVA from 35 h to less than 2 h for a model with more than104 reactions.

Matlab functions are available in the Supplementary Material or at https://github.com/maranasgroup/lll-FVA

Supplementary data are available at Bioinformatics online.