Biology subjects

Lazic, S. E.

Publications and source records attributed to Lazic, S. E..

7 recordsLinked to original sources

bioRxiv2020

Flexible fitting of PROTAC concentration-response curves with Gaussian Processes

A proteolysis targeting chimera (PROTAC) is a new technology that marks proteins for degradation in a highly specific manner. During screening, PROTAC compounds are tested in concentration-response (CR) assays to determine their potency, and parameters such as the half-maximal degradation concentration (DC50) are estimated from the fitted CR curves. These parameters are used to rank compounds, with lower DC50 values indicating greater potency. However, PROTAC data often exhibit bi-phasic and poly-phasic relationships, making standard sigmoidal CR models inappropriate. A common solution includes manual omitting of points (the so called "masking" step) allowing standard models to be used on the reduced datasets. Due to its manual and subjective nature, masking becomes a costly and non-reproducible procedure. We, therefore, used a Bayesian changepoint Gaussian Processes model that can flexibly fit both non-sigmoidal and sigmoidal CR curves without user input. Parameters, such as the DC50, the maximum effect Dmax, and the point of departure (PoD) are estimated from the fitted curves. We then rank compounds based on one or more parameters, and propagate the parameter uncertainty into the rankings, enabling us to confidently state if one compound is better than another. Hence, we used a flexible and automated procedure for PROTAC screening experiments. By minimizing subjective decisions, our approach reduces time, cost, and ensures reproducibility of the compound ranking procedure. The code and data are provided on GitHub (https://github.com/elizavetase-menova/gp_concentration_response).

cell biology↗

bioRxiv2020

Improving drug safety predictions by reducing poor analytical practices

Predicting the safety of a drug from preclinical data is a major challenge in drug discovery, and progressing an unsafe compound into the clinic puts patients at risk and wastes resources. In drug safety pharmacology and related fields, methods and analytical decisions known to provide poor predictions are common and include creating arbitrary thresholds, binning continuous values, giving all assays equal weight, and multiple reuse of information. In addition, the metrics used to evaluate models often omit important criteria and models performance on new data are often not assessed rigorously. Prediction models with these problems are unlikely to perform well, and published models suffer from many of these issues. We describe these problems in detail, demonstrate their negative consequences, and propose simple solutions that are standard in other disciplines where predictive modelling is used.

pharmacology and toxicology↗

bioRxiv2020

A Bayesian neural network for toxicity prediction

Predicting the toxicity of a compound preclinically enables better decision making, thereby reducing development costs and increasing patient safety. It is a complex issue, but in vitro assays and physico-chemical properties of compounds can be used to predict clinical toxicity. Neural networks (NNs) are a popular predictive tool due to their flexibility and ability to model non-linearities, but they are prone to overfitting and therefore are not recommended for small data sets. Furthermore, they dont quantify uncertainty in the predictions. Bayesian neural networks (BNNs) are able to avoid these pitfalls by using prior distributions on the parameters of a NN model and representing uncertainty about the predictions in the form of a distribution. We model the severity of drug-induced liver injury (DILI) to provide an example of a BNN performing better than a traditional but less flexible proportional odds logistic regression (POLR) model. We use appropriate metrics to evaluate predictions of the ordinal data type. To demonstrate the effect of a hierarchical prior for BNNs as an alternative to hyperparameter optimisation for NNs, we compare the performance of a BNN against NNs with dropout or penalty regularisation. We reduce the task to multiclass classification in order to be able to perform this comparison. A BNN trained for the multiclass classification produces poorer results than a BNN that captures the order. The current work lays a foundation for more complex models built on larger datasets, but can already be adopted by safety pharmacologists for risk quantification.

pharmacology and toxicology↗

bioRxiv2019

A Bayesian predictive approach for dealing with pseudoreplication

Pseudoreplication occurs when the number of measured values or data points exceeds the number of genuine replicates, and when the statistical analysis treats all data points as independent and thus fully contributing to the result. By artificially inflating the sample size, pseudoreplication contributes to irreproducibility, and it is a pervasive problem in biological research. In some fields, more than half of published experiments have pseudoreplication - making it one of the biggest threats to inferential validity. Researchers may be reluctant to use appropriate statistical methods if their hypothesis is about the pseudoreplicates and not the genuine replicates; for example, when an intervention is applied to pregnant female rodents (genuine replicates) but the hypothesis is about the effect on the multiple offspring (pseudoreplicates). We propose using a Bayesian predictive approach, which enables researchers to make valid inferences about biological entities of interest, even if they are pseudoreplicates, and show the benefits of this approach using two in vivo data sets.

neuroscience↗

bioRxiv2019

Determining organ weight toxicity with Bayesian causal models: Improving on the analysis of relative organ weights

Regulatory authorities require animal toxicity tests for new chemical entities. Organ weight changes are accepted as a sensitive indicator of chemically induced organ damage, but can be difficult to interpret because changes in organ weight might reflect chemically-induced changes in overall body weight. A common solution is to calculate the relative organ weight (organ to body weight ratio), but this inadequately controls for the dependence on body weight - a point made by statisticians for decades, but which has not been widely adopted. The recommended solution is an analysis of covariance (ANCOVA), but it is rarely used, possibly because both the method of statistical correction and the interpretation of the output may be unclear to those with minimal statistical training. Using relative organ weights can easily lead to incorrect conclusions, resulting in poor decisions, wasted resources, and an ethically questionable use of animals. We propose to cast the problem into a causal modelling framework as it directly assesses questions of scientific interest, the results are easy to interpret, and the analysis is simple to perform with freely available software. Furthermore, by taking a Bayesian approach we can model unequal variances, control for multiple testing, and directly provide evidence of safety.

pharmacology and toxicology↗

bioRxiv2017

Predicting drug safety and communicating risk: benefits of a Bayesian approach

Drug toxicity is a major source of attrition in drug discovery and development. Pharmaceutical companies routinely use preclinical data to predict clinical outcomes and continue to invest in new assays to improve predictions. However, there are many open questions about how to make the best use of available data, combine diverse data, quantify risk, and communicate risk and uncertainty to enable good decisions. The costs of suboptimal decisions are clear: resources are wasted and patients may be put at risk. We argue that Bayesian methods provide answers to all of these problems and use hERG-mediated QT prolongation as a case study. Benefits of Bayesian machine learning models include intuitive probabilistic statements of risk that incorporate all sources of uncertainty, the option to include diverse data and external information, and visualisations that have a clear link between the output from a statistical model and what this means for risk. Furthermore, Bayesian methods are easy to use with modern software, making their adoption for safety screening straightforward. We include R and Python code to encourage the adoption of these methods.

pharmacology and toxicology↗

bioRxiv2017

What exactly is ‘N’ in cell culture and animal experiments?

Biologists establish the existence of experimental effects by applying treatments or interventions to biological entities or units, such as people, animals, slice preparations, or cells. When done appropriately, independent replication of the entity-intervention pair contributes to the sample size (N) and forms the basis of statistical inference. However, sometimes the appropriate entity-intervention pair may not be obvious, and the wrong choice can make an experiment worthless. We surveyed a random sample of published animal experiments from 2011 to 2016 where interventions were applied to parents but effects examined in the offspring, as regulatory authorities have provided clear guidelines on replication with such designs. We found that only 22% of studies (95% CI = 17% to 29%) replicated the correct entity-intervention pair and thus made valid statistical inferences. Approximately half of the studies (46%, 95% CI = 38% to 53%) had pseudoreplication while 32% (95% CI = 26% to 39%) provided insufficient information to make a judgement. Pseudoreplication artificially inflates the sample size, leading to more false positive results and inflating the apparent evidence supporting a scientific claim. It is hard for science to advance when so many experiments are poorly designed and analysed. We argue that distinguishing between biological units, experimental units, and observational units clarifies where replication should occur, describe the criteria for genuine replication, and provide guidelines for designing and analysing in vitro, ex vivo, and in vivo experiments.

neuroscience↗