Quantitative Biology

We report on a method that classifies heart beats according to a set of 13 classes, including cardiac arrhythmias. The method localises the QRS peak complex to define each heart beat and uses a neural network to infer the patterns characteristic of each heart beat class. The best performing neural network contains six one-dimensional convolutional layers and four dense layers, with the kernel sizes being multiples of the characteristic scale of the problem, thus resulting a computationally fast and physically motivated neural network. For the same number of heart beat classes, our method yields better results with a considerably smaller neural network than previously published methods, which renders our method competitive for deployment in an internet-of-things solution.

Many studies are devoted to the design of radiomic models for a prediction task. When no effective model is found, it is often difficult to know whether the radiomic features do not include information relevant to the task or because of insufficient data. We propose a downsampling method to answer that question when considering a classification task into two groups. Using two large patient cohorts, several experimental configurations involving different numbers of patients were created. Univariate or multivariate radiomic models were designed from each configuration. Their performance as reflected by the Youden index (YI) and Area Under the receiver operating characteristic Curve (AUC) was compared to the stable performance obtained with the highest number of patients. A downsampling method is described to predict the YI and AUC achievable with a large number of patients. Using the multivariate models involving machine learning, YI and AUC increased with the number of patients while they decreased for univariate models. The downsampling method better estimated YI and AUC obtained with the largest number of patients than the YI and AUC obtained using the number of available patients and identifies the lack of information relevant to the classification task when no such information exists.

Prediction of protein structures using computational approaches has been explored for over two decades, paving a way for more focused research and development of algorithms in comparative modelling, ab intio modelling and structure refinement protocols. A tremendous success has been witnessed in template-based modelling protocols, whereas strategies that involve template-free modelling still lag behind, specifically for larger proteins (> 150 a.a.). Various improvements have been observed in ab initio protein structure prediction methodologies overtime, with recent ones attributed to the usage of deep learning approaches to construct protein backbone structure from its amino acid sequence. This review highlights the major strategies undertaken for template-free modelling of protein structures while discussing few tools developed under each strategy. It will also briefly comment on the progress observed in the field of ab initio modelling of proteins over the course of time as seen through the evolution of CASP platform.

Thanks to advancements in diagnosis and treatment, prostate cancer patients have high long-term survival rates. Currently, an important goal is to preserve quality-of-life during and after treatment. The relationship between the radiation a patient receives and the subsequent side effects he experiences is complex and difficult to model or predict. Here, we use machine learning algorithms and statistical models to explore the connection between radiation treatment and post-treatment gastro-urinary function. Since only a limited number of patient datasets are currently available, we used image flipping and curvature-based interpolation methods to generate more data in order to leverage transfer learning. Using interpolated and augmented data, we trained a convolutional autoencoder network to obtain near-optimal starting points for the weights. A convolutional neural network then analyzed the relationship between patient-reported quality-of-life and radiation. We also used analysis of variance and logistic regression to explore organ sensitivity to radiation and develop dosage thresholds for each organ region. Our findings show no connection between the bladder and quality-of-life scores. However, we found a connection between radiation applied to posterior and anterior rectal regions to changes in quality-of-life. Finally, we estimated radiation therapy dosage thresholds for each organ. Our analysis connects machine learning methods with organ sensitivity, thus providing a framework for informing cancer patient care using patient reported quality-of-life metrics.

Collision sports athletes experience many head acceleration events (HAEs) per season. The effects of these subconcussive events are largely understudied since HAEs may produce no overt symptoms, and are likely to diffusely manifest across multiple scales of study (e.g., molecular, cellular network, and behavior). This study integrated resting-state fMRI with metabolome, transcriptome and computational virtual reality (VR) behavior measures to assess the effects of exposure to HAEs on players in a collegiate American football team. Permutation-based mediation and moderation analysis was used to investigate relationships between network fingerprint, changes in omic measures and VR metrics over the season. Change in an energy cycle fatty acid, tridecenedioate, moderated the relationship between 1) miR-505 and DMN fingerprint and 2) the relationship between DMN fingerprint and worsening VR Balance measures (all p less than or equal to 0.05). In addition, the similarity in DMN over the season was negatively related to cumulative number of HAEs above 80G, and DMN fingerprint was less similar across the season in athletes relative to age-matched non-athletes. miR-505 was also positively related to average number of HAEs above 25G per session. It is important to note that tridecenedioate has a double bond making it a candidate for ROS scavenging. These findings between a candidate ROS-related metabolite, inflammatory miRNA, altered brain imaging and diminished behavioral performance suggests that impact athletes may experience chronic neuroinflammation. The rigorous permutation-based mediation/moderation may provide a methodology for investigating complex multi-scale biological data within humans alone and thus assist study of other functional brain problems.

Brain injuries are a major reason for mortality and morbidity following trauma in sports, work and traffic. Apart from the trauma at the site of impact (coup injury), other regions of the brain remote from the impact locations (non-coup) are commonly affected. We show that a screw theory-based method can be used to account for the combined effect of head rotational and linear accelerations in causing brain injuries. A scalar measure obtained from the inner product of the motion screw and the impact screw is shown to be a predictor of the severity and the location of non-coup brain injuries under an impact. The predictions are consistent with head impact experiments conducted with non-human primates. The methodology is proved using finite element simulations and already published experimental results

We propose a new method to estimate the time-varying effective (or instantaneous) reproduction number of the novel coronavirus disease (COVID-19). The method is based on a discrete-time stochastic augmented compartmental model that describes the virus transmission. A two-stage estimation method, which combines the Extended Kalman Filter (EKF) to estimate reported state variables (active and removed cases) and a low pass filter based on a rational transfer function to remove short term fluctuations of the reported cases, is used with case uncertainties that are assumed to follow a Gaussian distribution. Our method does not require information regarding serial intervals, which makes the estimation procedure simpler without reducing the quality of the estimate. We show that the proposed method is comparable to common approaches, e.g., age-structured and new cases based sequential Bayesian models. We also apply it to COVID-19 cases in the Scandinavian countries: Denmark, Sweden, and Norway, where we see a delay of about four days in predicting the epidemic peak.

Joint Species Distribution models (jSDMs) explain spatial variation in community composition by contributions of the environment, biotic associations, and possibly spatially structured residual variance. They show great promise as a general analytical framework for community ecology and macroecology, but current jSDMs scale poorly on large datasets, limiting their usefulness for novel community data, such as datasets generated using metabarcoding and metagenomics. Here, we present sjSDM, a novel method for estimating jSDMs that is based on Monte-Carlo integration of the joint likelihood. Implemented in PyTorch, a modern machine learning framework that can make use of CPU and GPU calculations, this approach is orders of magnitude faster than existing jSDM algorithms and can be scaled to very large datasets. Despite the dramatically improved speed, sjSDM produces the same predictive error and more accurate estimates of species association structures than alternative jSDM implementations. We provide our method in an R package to facilitate its applicability for practical data analysis.

Machine learning methods are increasingly employed to address challenges faced by biologists. One area that will greatly benefit from this cross-pollination is the problem of biological sequence design, which has massive potential for therapeutic applications. However, significant inefficiencies remain in communication between these fields which result in biologists finding the progress in machine learning inaccessible, and hinder machine learning scientists from contributing to impactful problems in bioengineering. Sequence design can be seen as a search process on a discrete, high-dimensional space, where each sequence is associated with a function. This sequence-to-function map is known as a "Fitness Landscape". Designing a sequence with a particular function is hence a matter of "discovering" such a (often rare) sequence within this space. Today we can build predictive models with good interpolation ability due to impressive progress in the synthesis and testing of biological sequences in large numbers, which enables model training and validation. However, it often remains a challenge to find useful sequences with the properties that we like using these models. In particular, in this primer we highlight that algorithms for experimental design, what we call "exploration strategies", are a related, yet distinct problem from building good models of sequence-to-function maps. We review advances and insights from current literature -- by no means a complete treatment -- while highlighting desirable features of optimal model-guided exploration, and cover potential pitfalls drawn from our own experience. This primer can serve as a starting point for researchers from different domains that are interested in the problem of searching a sequence space with a model, but are perhaps unaware of approaches that originate outside their field.

In recent years, cell-based assays have been frequently used in molecular interaction analysis. Cell-based assays complement traditional biochemical and biophysical methods, as they allow for molecular interaction analysis, mode of action studies and even drug screening processes to be performed under physiologically relevant conditions. In most cellular assays, biomolecules are usually labeled to achieve specificity. In order to overcome some of the drawbacks associated with label-based assays, we have recently introduced cell-based molography as a biosensor for the analysis of specific molecular interactions involving native membrane receptors in living cells. Here, we expand this assay to cytosolic protein-protein interactions. First, we created a biomimetic membrane receptor by tethering one cytosolic interaction partner to the plasma membrane. The artificial construct is then coherently arranged into a two-dimensional pattern within the cytosol of living cells. Thanks to the molographic sensor, the specific interactions between the coherently arranged protein and its endogenous interaction partners become visible in real-time without the use of a fluorescent label. This method turns out to be an important extension of cell-based molography because it expands the range of interactions that can be analyzed by molography to those in the cytosol of living cells.