Biology subjects

Spjuth, O.

Publications and source records attributed to Spjuth, O..

At least 19 recordsLinked to original sources

bioRxiv2026

Human-supervised Agentic AI for Hypothesis Generation and Experimental Assistance in Drug Repurposing

Computational drug repurposing has largely been focused on rapid hypothesis generation, yet real-world applications span a far broader lifecycle, from drug candidate suggestion to designing experiments, analyzing assay data, and iteratively refining candidates. Here, we demonstrate that agentic AI can operate throughout this lifecycle. To this end, we developed RepurAgent, a hierarchical multi-agent AI system comprising a supervisor agent and a planning agent that coordinate four specialized sub-agents (research, prediction, data, and report), through a human-in-the-loop design, with episodic memory and retrieval-augmented generation. The system is grounded in data, tools, and standard operating procedures specific for drug repurposing, developed within the REMEDi4ALL consortium. We validated the agentic system across three scenarios spanning the various stages within the repurposing lifecycle: in Acute Myeloid Leukemia, a blinded expert evaluation indicated that RepurAgent produced substantially more novel and mechanistically credible candidates compared to a vanilla LLM baseline; in a retrospective COVID-19 antiviral screen, RepurAgent acted as an adaptive experimental collaborator, prioritizing compounds with AUC-ROC up to 0.99 without predefined thresholds and flagging confounders missed in manual review; and for Multiple Sulfatase Deficiency, it prioritized 81 high-confidence candidates from 5000 compounds, which were further corroborated by domain experts. These results demonstrate that agentic AI can support across the drug repurposing lifecycle, from hypothesis generation to experimental analysis. RepurAgent is open source and deployed at https://repuragent.serve.scilifelab.se/.

bioinformatics↗

bioRxiv2025

Morphological cell profiling for drug repurposing against SARS-CoV-2 infection

Antiviral drug discovery has traditionally focused on targeting viral proteins, while host-directed strategies remain largely underexplored. Here, present a systematic drug repurposing strategy leveraging morphological profiling to identify host-targeting antivirals. Our image-based approach combines viral protein immunostaining with high-content Cell Painting analysis to simultaneously assess viral replication and provide in-depth analysis of host cell responses. By screening 5,275 repurposable drugs against SARS-CoV-2 infected cells, we identified compounds that reversed the infected cell phenotype, including ones not detected by conventional cytopathicity and antibody-based assays. A counter-screen excluded compounds whose antiviral activity was likely driven by drug-induced phospholipidosis (DIPL). Pathway enrichment analysis of compounds validated by both Cell Painting dose-response and DIPL assays, revealed host processes frequently hijacked by viruses, including innate immune responses and kinases. Among the top hits, both novel candidates, such as serdemetan, and previously reported broad-spectrum antivirals, such as sunitinib, were identified. Our approach constitutes an adaptable and scalable platform suited for diverse viral pathogens and cell systems. We provide a resource of open access screening data, images, and automated analysis pipelines to advance both antiviral discovery and pandemic preparedness.

microbiology↗

bioRxiv2025

Integrating Cell Painting and Thermal Proteome Profiling for Inference of Targets and Mechanism of Action

Understanding the mechanism of action (MoA) of bioactive compounds is a central challenge in drug discovery and chemical biology. In this article, we propose a strategy for integrating morphological data with proteomics to provide deeper insights into the MoA of compounds. We combine the rich phenotypic profiles of Cell Painting (CP) with the unbiased protein target detection of Thermal Proteome Profiling (TPP), and construct protein-protein interaction networks based on potential targets identified using both assays. We evaluate our method on public TPP data sets for the five compounds (+)-JQ1, I-BET151, Vemurafenib, Crizotinib and Panobinostat, together with Cell Painting data for 5270 drugs on U2OS cells. We show that the combined approach could accurately identify known MoAs for four out of five compounds. This work highlights the value of multimodal profiling for chemical biology and opens new avenues for data-driven discovery of therapeutic mechanisms.

bioinformatics↗

bioRxiv2025

Small molecule bioactivity benchmarks are often well-predicted by counting cells

Phenotypic profiling methods, such as Cell Painting and gene expression, have been widely used to predict compound bioactivity, often showing improvement over predictive models based on chemical structures alone. We discovered that a large subset of assays in widely-used benchmark datasets either directly relate to cell health and cytotoxicity or are assays intending to capture a more specific phenotype but whose active compounds impact cell count, while inactives do not. As a result, counting cells can achieve similar predictive performance as Cell Painting or gene expression data. Filtering benchmarks to include only assays relating to protein targets reveals that Cell Painting can capture information that cannot be predicted by mere cell counting. We re-evaluated three benchmark datasets used with Cell Painting data and observed that, in many cases, cell count models produced an AUC comparable to models using the full Cell Painting profiles. However, in protein-target-specific benchmarks across 17 distinct protein targets, Cell Painting features demonstrated unique predictive power, outperforming mean balanced accuracy from cell count models with a relative improvement of 19.6%. We propose five practical recommendations for benchmarking machine learning models for predicting bioactivity, including using cell count as a baseline feature. Although multi-class classification applications (such as matching samples based on their morphological profile) are less likely to be predictable by cell count than bioactivity benchmarks, these recommendations are broadly applicable to machine learning for drug discovery.

bioinformatics↗

bioRxiv2025

Label-Free Live-Cell Imaging improves Mode of Action Classification

Morphological profiling is a common approach to investigate the modes of action (MOAs) of compounds. Most methods rely on fixed-cell assays, which provide only a single snapshot at a predefined time point and overlook the dynamic nature of cellular responses. In contrast, live-cell imaging tracks responses over time, offering deeper insight into compound-specific effects and mechanisms; however, time-series analysis of image data remains challenging due to limited analytical tools. We present Live Cell Temporal Profiling (LCTP), a workflow for morphological profiling of label-free live-cell time series data that yields interpretable, biologically relevant results. We showcase LCTP in an MOA classification study using label-free data. The workflow integrates established deep-learning components, cell segmentation, live/dead classification, and single-cell feature extraction, with data-driven models to capture MOA-specific temporal phenotypes and produce time-resolved profiles that can be compared across compounds and cell lines. We assess MOA classification performance using double-blinded cross-validation simulating a real-world screening scenario. LCTP significantly improves MOA classification over single-time point analysis, consistently across both cell lines used in the study. Time-resolved phenotypic modeling reveals transient, sustained, and delayed responses, clarifying compound-specific temporal effects and mechanisms across MOAs. The presented workflow is modular: each step removes irrelevant information, enriching signal, and enabling straightforward updates as technologies evolve and as new technologies become available, while supporting reuse across studies broadly. We believe LCTP adds substantial value to high-throughput compound screening, showing that live-cell imaging combined with this workflow yields informative visualizations of temporal effects and improved MOA classification.

bioinformatics↗

bioRxiv2025

High-content morphological profiling by Cell Painting in 3D spheroids.

Cell Painting is a popular assay for morphological profiling of multi-labeled 2D monolayer cell cultures used in a wide range of applications. Culturing cells in 3D has potential for higher physiological relevance, such as when studying effects of perturbations. Robust and scalable 3D models can be challenging to characterize through imaging - particularly because light has difficulty penetrating cell multilayers. We introduce a scalable method where the Cell Painting assay is combined with tissue-clearing and applied to 3D spheroids generated in a ULA microplate format. Multi-channel images are acquired using confocal microscopy, and cells can be segmented inside those spheroids allowing for relevant morphological features to be extracted. Our end-to-end analysis pipeline comprises cell segmentation, morphological feature extraction, and between-spheroids and within-spheroid normalization. We demonstrate the method using spheroids cultured from two colorectal cancer cell lines and successfully detect distinct phenotypic changes upon compound treatments, on both spheroid-level using maximum intensity projections and on single cell-level. We show that drugs group by mechanism of action, with biologically relevant clusters especially evident with single-cell data. Finally, we contrast our method to results from 2D Cell Painting and discover a different pattern in DNA damaging drugs in HCT116 colorectal cancer cells. This work lays the foundation for multi-channel image-based screening in 3D spheroids.

pharmacology and toxicology↗

bioRxiv2025

Single-cell morphological profiling reveals insights into cell death

Analysis at the single-cell level is a powerful approach to study biological processes and responses to perturbations. However, its application in morphological profiling with phenomics remains underexplored. Here, we use the Cell Painting assay to investigate morphological effects of 53 small molecule compounds, associated with six distinct programmed cell death mechanisms, across six concentrations in MCF7 cells. To compare single-cell and aggregated analysis strategies, we conduct both supervised and unsupervised evaluations aimed at identifying features linked to programmed cell death. We apply an energy distance as a metric to quantify morphological perturbation strength, enabling efficient filtering. Among three tested feature extraction methods, self-supervised DINO embeddings applied to single-cell data captured high-resolution morphological patterns. Focused analyses of apoptosis-inducing compounds revealed biological heterogeneity attributable to specific molecular targets and concentration-dependent effects, which were not apparent in aggregated profiles. In contrast, multi-class classification models for the six programmed cell death mechanisms trained on single-cell features achieved F1 scores of 79.86%, while models trained on aggregated features reached F1 scores of up to 89.97%. Our results highlight the advantages of single-cell data for unsupervised exploration and show that aggregated representations yield more robust and accurate performance in supervised models.

bioinformatics↗

bioRxiv2024

Comparing Lipinskis Rule of 5 and Machine Learning Based Prediction of Fraction Absorbed for Assessing Oral Absorption in Humans

BackgroundThe influential Lipinskis Rule of 5 (Ro5) describes molecular properties important for oral absorption in humans. According to Ro5, poor absorption is more likely when 2 or more of its criteria (molecular weight (MW) >500 g/mol, calculated octanol-water partition coefficient (clog P) >5, >5 hydrogen bond donors (HBD) and >10 hydrogen bond acceptors (HBA)) are violated. Earlier evaluations have shown that many drugs are sufficiently well absorbed into the systemic circulation despite many Ro5-violations. No evaluation of Ro5 vs fraction absorbed (fa) has, however, been done. MethodsDatasets of orally administered drugs violating and not violating Ro5 and with available human clinical fa-values were assembled, and contrasted to machine learning based predictions using the ANDROMEDA prediction software having a major MW-domain of 150-750 g/mol. Results129 Ro5-violent compounds (29 with MW>1000 g/mol) were found, 59 of which had fa-values (42 % mean fa). 34 % and 66 % of compounds were predicted as having fa[≤]10% and >10-30 % respectively, which was in good agreement with measured fa of 37 % and 63 %. The fa for all compounds with fa[≤]5 and [≤]10 % were correctly predicted. For compounds with fa>30 %, 81 % were predicted to have a fa>30 %, but none were predicted to have a fa<10 %. The Q2 for predicted vs observed fa was 0.64. For a set of 77 compounds without Ro5 violation (80 % mean fa), all compounds were correctly predicted to have a fa below or above 30 % (Q2=0.56). Among these are compounds with poor uptake (<1 % to 7 %). ConclusionWe show that machine learning based predictions of fa are superior to Ro5 for assessing oral absorption obstacles in humans. Too strict reliance on Ro5 may hence constitute a risk. ANDROMEDA predicts fa well, easily and quickly, and also differentiates well between poor and adequate oral uptake for compounds violating and not-violating Ro5. This makes it a valid and useful tool capable of predicting oral absorption in humans with good accuracy and replacing Ro5 for oral absorption assessments.

pharmacology and toxicology↗

bioRxiv2024

A Decade in a Systematic Review: The Evolution and Impact of Cell Painting

High-content image-based assays have fueled significant discoveries in the life sciences in the past decade (2013-2023), including novel insights into disease etiology, mechanism of action, new therapeutics, and toxicology predictions. Here, we systematically review the substantial methodological advancements and applications of Cell Painting. Advancements include improvements in the Cell Painting protocol, assay adaptations for different types of perturbations and applications, and improved methodologies for feature extraction, quality control, and batch effect correction. Moreover, machine learning methods recently surpassed classical approaches in their ability to extract biologically useful information from Cell Painting images. Cell Painting data have been used alone or in combination with other - omics data to decipher the mechanism of action of a compound, its toxicity profile, and many other biological effects. Overall, key methodological advances have expanded Cell Paintings ability to capture cellular responses to various perturbations. Future advances will likely lie in advancing computational and experimental techniques, developing new publicly available datasets, and integrating them with other high-content data types.

cell biology↗

bioRxiv2024

Federated Learning for Predicting Compound Mechanism of Action Based on Image-data from Cell Painting

Having access to sufficient data is essential in order to train accurate machine learning models, but much data is not publicly available. In drug discovery this is particularly evident, as much data is withheld at pharmaceutical companies for various reasons. Federated Learning (FL) aims at training a joint model between multiple parties but without disclosing data between the parties. In this work, we leverage Federated Learning to predict compound Mechanism of Action (MoA) using fluorescence image data from cell painting. Our study evaluates the effectiveness and efficiency of FL, comparing to non-collaborative and data-sharing collaborative learning in diverse scenarios. Specifically, we investigate the impact of data heterogeneity across participants on MoA prediction, an essential concern in real-life applications of FL, and demonstrate the benefits for all involved parties. This work highlights the potential of federated learning in multi-institutional collaborative machine learning for drug discovery and assessment of chemicals, offering a promising avenue to overcome data-sharing constraints.

bioinformatics↗

bioRxiv2024

PKSmart: An Open-Source Computational Model to Predict in vivo Pharmacokinetics of Small Molecules

Drug exposure is a key contributor to the safety and efficacy of drugs. It can be defined using human pharmacokinetics (PK) parameters that affect the blood concentration profile of a drug, such as steady-state volume of distribution (VDss), total body clearance (CL), half-life (t[1/2]), fraction unbound in plasma (fu) and mean residence time (MRT). In this work, we used molecular structural fingerprints, physicochemical properties, and predicted animal PK data as features to model the human PK parameters VDss, CL, t[1/2], fu and MRT for 1,283 unique compounds. First, we predicted animal PK parameters [VDss, CL, fu] for rats, dogs, and monkeys for 372 unique compounds using molecular structural fingerprints and physicochemical properties. Second, we used Morgan fingerprints, Mordred descriptors and predicted animal PK parameters in a hyperparameter-optimised Random Forest algorithm to predict human PK parameters. When validated using repeated nested cross-validation, human VDss was best predicted with an R2 of 0.55 and a Geometric Mean Fold Error (GMFE) of 2.09; CL with accuracies of R2=0.31 and GMFE=2.43, fu with R2=0.61 and GMFE=2.81, MRT with R2=0.28 and GMFE=2.49, and t[1/2] with R2=0.31 and GMFE=2.46 for models combining Morgan fingerprints, Mordred descriptors and predicted animal PK parameters. We evaluated models with an external test set comprising 315 compounds for VDss (R2=0.33 and GMFE=2.58) and CL (R2=0.45 and GMFE=1.98). We compared our models with proprietary pharmacokinetic (PK) models from AstraZeneca and found that model predictions were similar with Pearson correlations ranging from 0.77-0.78 for human PK parameters of VDss and fu and 0.46-0.71 for animal (dog and rat) PK parameters of VDss, CL and fu. To the best of our knowledge, this is the first work that publicly releases PK models on par with industry-standard models. Early assessment and integration of predicted PK properties are crucial, such as in DMTA cycles, which is possible with models in this study based on the input of only chemical structures. We developed a webhosted application PKSmart (https://broad.io/PKSmart) which users can access using a web browser with all code also downloadable for local use. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=73 SRC="FIGDIR/small/578658v1_ufig1.gif" ALT="Figure 1"> View larger version (18K): org.highwire.dtl.DTLVardef@1a4b673org.highwire.dtl.DTLVardef@c64c5aorg.highwire.dtl.DTLVardef@bec0b3org.highwire.dtl.DTLVardef@15d5756_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOFigure:C_FLOATNO For TOC Only C_FIG

bioinformatics↗

bioRxiv2024

Improved Early Detection of Drug-Induced Liver Injury by Integrating Predicted in vivo and in vitro Data

Drug-induced liver injury (DILI) has been significant challenge in drug discovery, often leading to clinical trial failures and necessitating drug withdrawals. The existing suite of in vitro proxy-DILI assays is generally effective at identifying compounds with hepatotoxicity. However, there is considerable interest in enhancing in silico prediction of DILI because it allows for the evaluation of large sets of compounds more quickly and cost-effectively, particularly in the early stages of projects. In this study, we aim to study ML models for DILI prediction that first predicts nine proxy-DILI labels and then uses them as features in addition to chemical structural features to predict DILI. The features include in vitro (e.g., mitochondrial toxicity, bile salt export pump inhibition) data, in vivo (e.g., preclinical rat hepatotoxicity studies) data, pharmacokinetic parameters of maximum concentration, structural fingerprints, and physicochemical parameters. We trained DILI-prediction models on 888 compounds from the DILIst dataset and tested on a held-out external test set of 223 compounds from DILIst dataset. The best model, DILIPredictor, attained an AUC-ROC of 0.79. This model enabled the detection of top 25 toxic compounds compared to models using only structural features (2.68 LR+ score). Using feature interpretation from DILIPredictor, we were able to identify the chemical substructures causing DILI as well as differentiate cases DILI is caused by compounds in animals but not in humans. For example, DILIPredictor correctly recognized 2-butoxyethanol as non-toxic in humans despite its hepatotoxicity in mice models. Overall, the DILIPredictor model improves the detection of compounds causing DILI with an improved differentiation between animal and human sensitivity as well as the potential for mechanism evaluation. DILIPredictor is publicly available at https://broad.io/DILIPredictor for use via web interface and with all code available for download and local implementation via https://pypi.org/project/dilipred/. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=181 SRC="FIGDIR/small/575128v2_ufig1.gif" ALT="Figure 1"> View larger version (56K): org.highwire.dtl.DTLVardef@1b3346org.highwire.dtl.DTLVardef@1c3460dorg.highwire.dtl.DTLVardef@12c38cdorg.highwire.dtl.DTLVardef@455aee_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2023

CPSign - Conformal Prediction for Cheminformatics Modeling

Conformal prediction has seen many applications in pharmaceutical science, being able to calibrate outputs of machine learning models and producing valid prediction intervals. We here present the open source software CPSign that is a complete implementation of conformal prediction for cheminformatics modeling. CPSign implements inductive and transductive conformal prediction for classification and regression, and probabilistic prediction with the Venn-ABERS methodology. The main chemical representation is signatures but other types of descriptors are also supported. The main modeling methodology is support vector machines (SVMs), but additional modeling methods are supported via an extension mechanism, e.g. DeepLearning4j models. We also describe features for visualizing results from conformal models including calibration and efficiency plots, as well as features to publish predictive models as REST services. We compare CPSign against other common cheminformatics modeling approaches including random forest, and a directed message-passing neural network. The results show that CPSign produces robust predictive performance with comparative predictive efficiency, with superior runtime and lower hardware requirements compared to neural network based models. CPSign has been used in several studies and is in production-use in multiple organizations. The ability to work directly with chemical input files, perform descriptor calculation and modeling with SVM in the conformal prediction framework, with a single software package having a low footprint and fast execution time makes CPSign a convenient and yet flexible package for training, deploying, and predicting on chemical data.

bioinformatics↗

bioRxiv2023

Insights into Drug Cardiotoxicity from Biological and Chemical Data: The First Public Classifiers for FDA DICTrank

Drug-induced cardiotoxicity (DICT) is a major concern in drug development, accounting for 10-14% of postmarket withdrawals. In this study, we explored the capabilities of various chemical and biological data to predict cardiotoxicity, using the recently released Drug-Induced Cardiotoxicity Rank (DICTrank) dataset from the United States FDA. We analyzed a diverse set of data sources, including physicochemical properties, annotated mechanisms of action (MOA), Cell Painting, Gene Expression, and more, to identify indications of cardiotoxicity. We found that such data, including protein targets, especially those related to ion channels (such as hERG), physicochemical properties (such as electrotopological state) as well as peak concentration in plasma offer strong predictive ability as well as valuable insights into DICT. We also found compounds annotated with particular mechanisms of action, such as cyclooxygenase inhibition, could distinguish between most-concern and no-concern DICT compounds. Cell Painting features related to ER stress discern the most-concern cardiotoxic compounds from non-toxic compounds. While models based on physicochemical properties currently provide substantial predictive accuracy (AUCPR = 0.93), this study also underscores the potential benefits of incorporating more comprehensive biological data in future DICT predictive models. With the availability of - omics data in the future, using biological data promises enhanced predictability and delivers deeper mechanistic insights, paving the way for safer therapeutic drug development. All models and data used in this study are publicly released at https://broad.io/DICTrank_Predictor O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=136 SRC="FIGDIR/small/562398v1_ufig1.gif" ALT="Figure 1"> View larger version (32K): org.highwire.dtl.DTLVardef@755906org.highwire.dtl.DTLVardef@27c43borg.highwire.dtl.DTLVardef@c430dborg.highwire.dtl.DTLVardef@63b2ad_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2023

From Pixels to Phenotypes: Integrating Image-Based Profiling with Cell Health Data Improves Interpretability

Cell Painting assays generate morphological profiles that are versatile descriptors of biological systems and have been used to predict in vitro and in vivo drug effects. However, Cell Painting features are based on image statistics, and are, therefore, often not readily biologically interpretable. In this study, we introduce an approach that maps specific Cell Painting features into the BioMorph space using readouts from comprehensive Cell Health assays. We validated that the resulting BioMorph space effectively connected compounds not only with the morphological features associated with their bioactivity but with deeper insights into phenotypic characteristics and cellular processes associated with the given bioactivity. The BioMorph space revealed the mechanism of action for individual compounds, including dual-acting compounds such as emetine, an inhibitor of both protein synthesis and DNA replication. In summary, BioMorph space offers a more biologically relevant way to interpret cell morphological features from the Cell Painting assays and to generate hypotheses for experimental validation. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=163 SRC="FIGDIR/small/549031v1_ufig1.gif" ALT="Figure 1"> View larger version (33K): org.highwire.dtl.DTLVardef@1a6169forg.highwire.dtl.DTLVardef@11757eforg.highwire.dtl.DTLVardef@18e3fc2org.highwire.dtl.DTLVardef@1fc0a90_HPS_FORMAT_FIGEXP M_FIG C_FIG IN BRIEFSeal et al. used machine learning models and feature selection approaches to group cell morphological features from Cell Painting assays and to describe the shared role of these morphological features in various cell health phenotypes. The resulting BioMorph space improves the ability to understand the mechanism of action and toxicity of compounds and to generate hypotheses to guide future experiments. HIGHLIGHTSO_LICombining Cell Painting and Cell Health imaging data defines the BioMorph space. C_LIO_LIBioMorph space allows detecting less common mechanisms for bioactive compounds. C_LIO_LIBioMorph space can generate MOA hypotheses to guide experimental validation. C_LIO_LIBioMorph space is more biologically relevant and interpretable than Cell Painting features. C_LI

cell biology↗

bioRxiv2023

Prediction of the Human Pharmacokinetics of 30 Modern Antibiotics Using the ANDROMEDA Software

The ANDROMEDA software, based on machine learning, conformal prediction and a new physiologically-based pharmacokinetic model, was used to predict and characterize the human clinical pharmacokinetics of 30 selected modern small antibiotic compounds (investigational and marketed drugs). A majority of clinical pharmacokinetic data was missing. ANDROMEDA successfully filled this gap. Most antibiotics were predicted and measured to have limited permeability, good metabolic stability and multiple elimination pathways. According to predictions, most of the antibiotics are mainly eliminated renally and biliary and every other antibiotic is mainly eliminated via the renal route. Mean prediction errors for steady state volume of distribution, unbound fraction in plasma, renal and total clearance, oral clearance, fraction absorbed, fraction excreted renally, oral bioavailability and half-life were 1.3- to 2.3-fold. The overall median and maximum prediction errors were 1.5- and 4.8-fold, respectively, and 92 % of predictions had <3-fold error. Results are consistent with those obtained in previous validation studies and are better than with the best laboratory-based prediction methods, which validates ANDROMEDA for predictions of human clinical pharmacokinetics of modern antibiotic drugs, which to a great extent demonstrate pharmacokinetic characteristics challenging for laboratory methods (metabolic stability, limited permeability, efflux and multiple elimination pathways). Advantages with ANDROMEDA include that results are produced without the use of animals and cells and that predictions and decision-making can be done already at the design stage.

pharmacology and toxicology↗

bioRxiv2023

Application of the ANDROMEDA Software for Prediction of the Human Pharmacokinetics of Modern Anticancer Drugs

The ANDROMEDA toolkit for prediction of human clinical pharmacokinetics, based on machine learning, conformal prediction and a new physiologically-based pharmacokinetic model, was used to predict and characterize the human clinical pharmacokinetics of 12 small anticancer drugs marketed in 2021 and 2022 (molecular weight 355 to 1326 g/mol). The study is part of a series of software validations. A majority of clinical pharmacokinetic data was missing. ANDROMEDA successfully filled this gap. Most drugs were predicted/measured to have relatively complex pharmacokinetics, with limited passive permeability+efflux, high degree of plasma protein binding, significant gut-wall elimination and food interaction, biliary excretion and/or limited dissolution potential. Median, mean and maximum prediction errors for steady state volume of distribution, unbound fraction in plasma, blood-to-plasma concentration ratio, hepatic, renal and total clearance, fraction absorbed, oral bioavailability, half-life and degree of food interaction were 1.6-, 2.4- and 17-fold, respectively. Less than 3-fold errors were found for 78 % of predictions. Results are consistent with those obtained in previous validation studies and are better than with the best laboratory-based prediction methods, which validates ANDROMEDA for predictions of human clinical pharmacokinetics of modern small anticancer drugs with multi-mechanistical and challenging pharmacokinetics.

pharmacology and toxicology↗

bioRxiv2022

Predicting the Influence of Fat Food Intake on the Absorption and Systemic Exposure of Modern Small Drugs using ANDROMEDA by Prosilico Software

AO_SCPLOWBSTRACTC_SCPLOWO_ST_ABSIntroductionC_ST_ABSThe ANDROMEDA software by Prosilico has previously been successfully applied and validated for predictions of absorption characteristics of small drugs in man. The influence of fat food on the gastrointestinal uptake and systemic exposure of drugs have, however, not yet been evaluated with the software. Objective and MethodologyThe main objective was to use ANDROMEDA to predict area under the plasma concentration-time curve ratios in the fed (fat food) and fasted states (AUCfed/AUCfast) for small drugs (including those marketed in 2021) and compare results with corresponding measured clinical estimates. Actual dose sizes were considered. Another objective was to compare the performance of ANDROMEDA vs physiologically based pharmacokinetic (PBPK) modelling and simulations by The Food Effect PBPK IQ Working Group. PBPK results generated using Simcyp and GastroPlus software were based on various physicochemical, in vitro and in vivo data and a decision tree for model verification and optimization. Results and Discussion63 drugs, including 17 new drugs, with observed AUCfed/AUCfast between 0.2 and 5.5 were found and used for this evaluation. Predicted AUCfed/AUCfast had mean and maximum errors of 1.5- and 4.1-fold, respectively, and the predictive accuracy (correlation between predicted and observed AUCfed/AUCfast; Q2) was 0.3. 14 % of predictions had >2-fold error. For 72 % of drugs, food interaction class was correctly predicted. The level of predictive accuracy was overall similar to results obtained with PBPK modelling and simulations, however, with lower maximum error and higher compound coverage. With PBPK models, maximum simulation error was 7.7-fold and 3 highly lipophilic compounds were not possible to simulate. ConclusionThe results validate ANDROMEDA for prediction of fat food-drug interaction size for small drugs in man. Major advantages with the methodology include that prediction results are produced directly from molecular structure and oral dose and are similar to PBPK-simulation results obtained using in vitro and clinical data. Furthermore, ANDROMEDA showed lower maximum errors and wider compound range.

pharmacology and toxicology↗