Response Prediction Model Research Articles

Developing predictive models for tocilizumab response in rheumatoid arthritis: a gene expression and machine learning approaches

Background: Rheumatoid arthritis (RA) is a chronic autoimmune disease treated with tocilizumab in patients unresponsive to methotrexate. This study aimed to identify gene expression profiles and predictive models for tocilizumab treatment response in RA patients. Methods: Using the GSE78068 dataset from 38 RA patients, we identified differentially expressed genes between remission and non-remission groups. Predictive models were created using CART, random forest, and SVM techniques, with model genes selected through LASSO regression. Gene set enrichment and immune cell landscape analyses were performed to understand biological pathways and immune cell composition. Results: Analysis revealed 40 differentially expressed genes, with LASSO regression identifying 8 model genes significantly associated with remission. The SVM-based model achieved the highest performance (AUC=1.0, Brier score=0.025). Conclusions: This study developed an effective 8-gene model for predicting tocilizumab treatment response, potentially supporting personalized therapy in RA patients.

Oral Microbiota Associated with Clinical Efficacy of Ustekinumab in Crohn's Disease.

Crohn's Disease (CD) is a chronic inflammatory gastrointestinal disease. Ustekinumab (UST) has been utilized as a therapeutic option for CD patients. However, approximately 40-60% of patients exhibit an inadequate response to UST. Accumulating evidence has confirmed the involvement of oral bacteria in the development of CD. Nevertheless, the relationship between oral microbiota and the efficacy of UST therapy in CD patients has remained unexplored. We recruited 28 healthy individuals and 53 CD patients, 47 of whom completed the entire UST therapy. Oral samples and clinical data were collected. The clinical response and clinical remission were defined based on the CDAI score. Oral samples were analyzed by 16S rRNA gene sequencing. The analysis of sequence data was performed by QIIME and R. We revealed the oral microbial difference between the Healthy Control (HC) group and the CD group. The enrichment of Fusobacteria, Leptotrichia, Capnocytophaga, and Campylobacter, and the diminution of Haemophilus and Rothia were observed in the CD group. Differences in oral microbiota were also identified among patients with different efficacy of UST. Compared to the response and remission groups, both the non-response and non-remission groups showed significantly higher levels of Fusobacteria and Leptotrichia. Predictive models for clinical response and clinical remission in UST were developed based on oral microbiota, with the Area Under the Curve (AUC) value of 0.944 and 0.930, respectively. Oral microbiota was relevant to the UST efficacy in patients with CD based on the predictive model. These findings suggest that oral microbiota could serve as a non-invasive prognostic biomarker for UST treatment in CD patients.

Comparative analysis of regression algorithms for drug response prediction using GDSC dataset

BackgroundDrug response prediction can infer the relationship between an individual’s genetic profile and a drug, which can be used to determine the choice of treatment for an individual patient. Prediction of drug response is recently being performed using machine learning technology. However, high-throughput sequencing data produces thousands of features per patient. In addition, it is difficult for researchers to know which algorithm is appropriate for prediction as various regression and feature selection algorithms exist.MethodsWe compared and evaluated the performance of 13 representative regression algorithms using Genomics of Drug Sensitivity in Cancer (GDSC) dataset. Three analyses was conducted to show the effect of feature selection methods, multiomics information, and drug categories on drug response prediction.ResultsIn the experiments, Support Vector Regression algorithm and gene features selected with LINC L1000 dataset showed the best performance in terms of accuracy and execution time. However, integration of mutation and copy number variation information did not contribute to the prediction. Among the drug groups, responses of drugs related with hormone-related pathway were predicted with relatively high accuracy.ConclusionThis study can help bioinformatics researchers design data processing steps and select algorithms for drug response prediction, and develop a new drug response prediction model based on the GDSC or other high-throughput sequencing datasets.

Tailoring industrial enzymes for thermostability and activity evolution by the machine learning-based iCASE strategy

The pursuit of obtaining enzymes with high activity and stability remains a grail in enzyme evolution due to the stability-activity trade-off. Here, we develop an isothermal compressibility-assisted dynamic squeezing index perturbation engineering (iCASE) strategy to construct hierarchical modular networks for enzymes of varying complexity. Molecular mechanism analysis elucidates that the peak of adaptive evolution is reached through a structural response mechanism among variants. Furthermore, this dynamic response predictive model using structure-based supervised machine learning is established to predict enzyme function and fitness, demonstrating robust performance across different datasets and reliable prediction for epistasis. The universality of the iCASE strategy is validated by four sorts of enzymes with different structures and catalytic types. This machine learning-based iCASE strategy provides guidance for future research on the fitness evolution of enzymes.

Machine learning prediction model of the treatment response in schizophrenia reveals the importance of metabolic and subjective characteristics.

Predicting early treatment response in schizophrenia is pivotal for selecting the best therapeutic approach. Utilizing machine learning (ML) technique, we aimed to formulate a model predicting antipsychotic treatment outcomes. Data were obtained from 299 patients with schizophrenia from three multicenter, open-label, non-comparative clinical trials. For prediction of treatment response at weeks 4, 8, and 24, psychopathology (both objective and subjective symptoms), sociodemographic and clinical factors, functional outcomes, attitude toward medication, and metabolic characteristics were evaluated. Various ML techniques were applied. The highest area under the curve (AUC) at weeks 4, 8 and 24 was 0.711, 0.664 and 0.678 with extreme gradient boosting, respectively. Notably, our findings indicate that BMI and attitude toward medication play a pivotal role in predicting treatment responses at all-time points. Other salient features for weeks 4 and 8 included psychosocial functioning, negative symptoms, subjective symptoms like psychoticism and hostility, and the level of prolactin. For week 24, positive symptoms, depression, education level and duration of illness were also important. This study introduced a precise clinical model for predicting schizophrenia treatment outcomes using multiple readily accessible predictors. The findings underscore the significance of metabolic parameters and subjective traits.

Artificial Intelligence and Machine Learning Predicting Transarterial Chemoembolization Outcomes: A Systematic Review.

Major society guidelines recommend transarterial chemoembolization (TACE) as the standard of care for intermediate-stage hepatocellular carcinoma (HCC) patients. However, predicting treatment response remains challenging. As artificial intelligence (AI) may predict therapeutic responses, this systematic review aims to assess the performance and effectiveness of radiomics and AI-based models in predicting TACE outcomes in patients with HCC. A systemic search was conducted on Medline and Embase databases from inception to 7th April 2024. Included studies generated a predictive model for TACE response and evaluated its performance by area under the curve (AUC), specificity, or sensitivity analysis. Systematic reviews, meta-analyses, case series and reports, pediatric, and animal studies were excluded. Secondary search of references of included articles ensured comprehensiveness. 64 articles, with 13,412 TACE-treated patients, were included. AI in pre-treatment CT scans provided value in predicting the efficacy of TACE in HCC treatment. A positive association was observed for AI in pre-treatment MRI scans. Models incorporating radiomics had numerically better performance than those incorporating manual measured radiological variables. 39 studies demonstrated that combined predictive models had numerically better performance than models based solely on imaging or non-imaging features. A combined predictive model incorporating clinical features, laboratory investigations, and radiological characteristics may effectively predict response to TACE treatment for HCC.

A machine learning based prediction model for the impact mechanical response of composite laminates considering microstructure sensitive transverse properties

AbstractThe accurate and efficient prediction of impact mechanical response is crucial for safety design of composite structures. In this work, high‐fidelity representative volume elements (RVEs) with fiber, matrix and fiber/matrix interface are established, in which random fiber distributions are considered. A failure envelope under transverse loads is proposed based on computational micromechanical RVEs, and it is implemented by ABAQUS VUMAT subroutines to predict the mechanical response of composite laminates under impact loads. Based on a dataset from computational macromechanical finite element simulations, an artificial neural network model is established and trained. It is found that the random fiber distribution introduced a more obvious fluctuation to tension/compression strength than shear strength. The proposed failure criteria showed a better performance than Hashin and Tsai‐Wu criteria especially in combined compression and shear loads. An ANN model with 8 hidden layers can achieve the prediction with an acceptable coefficient of determination (R2) 0.98 and loss functions of mean absolute error (MAE) 71. For certain impact loading conditions, the well trained machine learning model predicted impact contact force history within 30 min, while the FEA costs about more than 75 min on the same computer. The prediction speed is increased by over 60% for certain impact loading conditions. It is hence shown that this method provides a potential alternative for evaluation of the impact resistance of composite structures.Highlights High‐fidelity computational micromechanics analysis based on representative volume elements are performed to uncover the complex relationship between microstructure and transverse strengths of composite laminates. The microstructure dependent transverse strength criterion shows high accuracy in combined compression and shear loads compared with Hashin and Tsai‐Wu criteria. A multi‐layer artificial neural network model is established and trained for the rapid prediction of impact force history for composite laminates. The rapid prediction method is achieved with a coefficient of determination of 0.98, and the prediction speed is increased by more than 60% for certain impact loading conditions.

Relationships between the Microbiome and Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer.

This study aimed to investigate the dynamic changes in the microbiome of patients with locally advanced rectal cancer (LARC) undergoing neoadjuvant chemoradiotherapy (nCRT), focusing on the relationship between the microbiome and response to nCRT. We conducted a longitudinal study involving 103 samples from 26 patients with LARC. Samples were collected from both the tumor and normal rectal tissues before and after nCRT. Diversity, taxonomic, and network analyses were performed to compare the microbiome profiles across different tissue types, pre- and post-nCRT time-points, and nCRT responses. Between the tumor and normal tissue samples, no differences in microbial diversity and composition were observed. However, when pre- and post-nCRT samples were compared, there was a significant decrease in diversity, along with notable changes in composition. Non-responders exhibited more extensive changes in their microbiome composition during nCRT, characterized by an increase in pathogenic microbes. Meanwhile, responders had relatively stable microbiome communities with more enriched butyrate-producing bacteria. Network analysis revealed distinct patterns of microbial interactions between responders and non-responders, where butyrate-producing bacteria formed strong networks in responders, while opportunistic pathogens formed strong networks in non-responders. A Bayesian network model for predicting the nCRT response was established, with butyrate-producing bacteria playing a major predictive role. Our study demonstrated a significant association between the microbiome and nCRT response in LARC patients, leading to the development of a microbiome-based response prediction model. These findings suggest potential applications of microbiome signatures for predicting and optimizing nCRT treatment in LARC patients.

A precision oncology-focused deep learning framework for personalized selection of cancer therapy.

Precision oncology matches tumors to targeted therapies based on the presence of actionable molecular alterations. However, most tumors lack actionable alterations, restricting treatment options to cytotoxic chemotherapies for which few data-driven prioritization strategies currently exist. Here, we report an integrated computational/experimental treatment selection approach applicable for both chemotherapies and targeted agents irrespective of actionable alterations. We generated functional drug response data on a large collection of patient-derived tumor models and used it to train ScreenDL, a novel deep learning-based cancer drug response prediction model. ScreenDL leverages the combination of tumor omic and functional drug screening data to predict the most efficacious treatments. We show that ScreenDL accurately predicts response to drugs with diverse mechanisms, outperforming existing methods and approved biomarkers. In our preclinical study, this approach achieved superior clinical benefit and objective response rates in breast cancer patient-derived xenografts, suggesting that testing ScreenDL in clinical trials may be warranted.

Growth Prediction Model for Prepubertal Children With Idiopathic Growth Hormone Deficiency: An Analysis of LG Growth Study Data.

Growth hormone (GH) treatment is effective in restoring normal growth in children with GH deficiency (GHD). However, individual responses to GH treatment vary, necessitating predictive models to estimate growth outcomes. This study aimed to develop and validate a predictive model for GH treatment response during the first 2 years in patients with idiopathic GHD using the LG growth study (LGS) database. This observational study included 669 prepubertal patients with idiopathic GHD from the LGS registry who received GH treatment for at least 2 years. Clinical and laboratory data were collected at baseline and every 6 months thereafter. Stepwise multivariate regression analysis was performed to develop prediction models for the treatment period. The mean age of patients with GDH was 6.0 ± 1.8 years. Height standard deviation score (SDS) significantly increased from -2.50 ± 0.71 to -1.66 ± 0.71 in the first year and -1.35 ± 0.71 in the second year. The first-year growth velocity was 9.06 ± 1.51 cm, decreasing to 7.42 ± 1.37 cm in the second year. The prediction models incorporated variables such as age, birth weight, bone age, initial height SDS, body mass index SDS, mid-parental height, GH dose and first year of height after GH treatment, explaining 76.9% and 84.1% of the variability in height SDS changes in the first and second years, respectively. GH treatment significantly improves height outcomes in prepubertal children with GHD. The developed predictive models demonstrated accuracy, facilitating personalized GH therapy. Future research should focus on refining these models and exploring the long-term effects of GH treatment in pubertal patients. ClinicalTrials.gov identifier: NCT01604395.

How accurate are machine learning models in predicting anti-seizure medication responses: A systematic review.

Current epilepsy management protocols often depend on anti-seizure medication (ASM) trials and assessment of clinical response. This may delay the initiation of the ASM regimen that might optimally balance efficacy and tolerability for individual patients. Machine learning (ML) can offer a promising tool for efficiently predicting ASM response. The objective of this review is to synthesize the available information about the effectiveness and limitations of ML models in predicting and classifying the response of patients with epilepsy to ASMs, and to assess the impact of various data inputs on prediction performance. We conducted a comprehensive search of studies utilizing ML models for ASM response prediction using PubMed and Scopus up until November 2024. The review included 37 studies. Various data types, including clinical information, brain MRI, EEG, and genetic data, are useful in predicting responses to ASMs. Tree-based ML algorithms and Support Vector Machines are the most used models. Reported results vary widely, with certain models achieving near-perfect accuracy and others performing similar to random classifiers. The review also highlights the limitations of this research field, especially concerning the quality and quantity of data. The findings indicate that while ML models show great promise in predicting ASM responses in epilepsy, further research is required to refine these models for practical clinical application. The review underscores both the potential of ML in advancing precision medicine in epilepsy management and the need for continued research to improve prediction accuracy.

P25 Genome-wide association study and predictive models for the efficacy and safety of NB-UVB phototherapy in psoriasis

Abstract Background Psoriasis is a chronic inflammatory skin disease affecting ∼125 million individuals worldwide. Although the treatment of psoriasis has entered an era of biological agents and small molecule drugs, narrowband ultraviolet B (NB-UVB) phototherapy still represents an effective, cost-efficient, and safe therapeutic option for patients with plaque psoriasis. However, the clinical and genetic predictors of NB-UVB treatment response remain limited. Objective We aimed to detect single nucleotide polymorphisms (SNPs) associated with the efficacy of NB-UVB in treating psoriasis and its adverse events (AEs), in order to establish predictive models for the efficacy and safety of NB-UVB phototherapy. Methods A total of 252 patients with moderate-to-severe plaque psoriasis were enrolled in this study and underwent 12 weeks of NB-UVB phototherapy. Psoriasis area and severity index (PASI) and the AEs of NB-UVB treatment were recorded during follow-up. Whole blood samples were collected from the participants, and extracted DNA was genotyped using Infinium Global Screening Array-24 v3.0 BeadChip. Genome-wide association studies (GWASs) were performed to identify SNPs related to NB-UVB treatment response. Predictive models for the efficacy and safety of NB-UVB therapy, which involved SNPs along with clinical parameters, were developed utilizing logistic regression and further validated by receiver operating characteristic (ROC) curve and calibration curve. Results After 4 weeks of NB-UVB phototherapy, 40.8% of patients achieved PASI50. Through GWASs, we discovered that SLC7A13 rs35314286 TA allele [odds ratio (OR) = 2.631, P = 5.96 × 10−6] and DYNC1H1 rs941636 A allele (OR = 3.066, P = 3.15 × 10−6) were significantly correlated with better efficacy of 4-week NB-UVB treatment. After 12 weeks of treatment, 61.4% of patients achieved PASI75. Seven ATP2B2 SNPs and 11 PSMB7 SNPs significantly impacted 12-week PASI75 achievement. During the treatment, 29.8% of patients experienced AEs that included pruritus, sting, numbness, erythema, vesicles, desquamation, hyperpigmentation, and photosensitivity. Two KCNA2 SNPs, seven THSD7B SNPs and one TENM4 SNP were considerably associated with the occurrence of AEs. Based on the clinical parameters (body mass index, smoking history, PASI score, and the percentage of PASI improvement at Week 4) and NB-UVB relevant SNPs, predictive models for the efficacy and safety of NB-UVB phototherapy were established, demonstrating excellent predictive capacities with area under the curve (AUC) value being 0.81 for 4-week efficacy, 0.80 for 12-week efficacy, and 0.85 for AE development. Conclusion Multiple gene SNPs were identified to be correlated with the efficacy and safety of NB-UVB phototherapy. Moreover, predictive models of NB-UVB treatment response were subsequently established. Our findings might contribute to the advancement of precision phototherapy in psoriasis. Clinical trial registration (www.chictr.org.cn), identifier (ChiCTR2000036186).

Dynamical analysis of the Rulkov model with quasiperiodic forcing

This paper investigates the dynamical behavior of the Rulkov neuron model under quasiperiodic forcing, focusing on the emergence and mechanisms of strange nonchaotic attractors (SNAs). SNAs are characterized by fractal geometry without sensitive dependence on initial conditions, and their formation mechanisms, such as torus-doubling, fractal, bubble, Heagy–Hammel, and multi-intermittency routes, are explored in detail. The study employs Lyapunov exponents, singular continuous spectrum, and phase sensitivity to detect and characterize the strange and nonchaotic properties of attractors. Numerical simulations reveal the transition dynamics between strange nonchaotic, quasiperiodic, and chaotic states, emphasizing the role of transient dynamics in understanding the complex behavior of neural systems under multifrequency stimuli. The findings provide deeper insights into the intricate dynamical processes in neuron models, potentially aiding the development of advanced models for neural response prediction in varying environments.

Multi-stain deep learning prediction model of treatment response in lupus nephritis based on renal histopathology.

The response of the kidney after induction treatment is one of the determinants of prognosis in lupus nephritis, but effective predictive tools are lacking. Here, we sought to apply deep learning approaches on kidney biopsies for treatment response prediction in lupus nephritis. Patients who received cyclophosphamide or mycophenolate mofetil as induction treatment were included and the primary outcome was 12-month treatment response, complete response defined as 24h urinary protein under 0.5 g with normal estimated glomerular filtration rate or within 10% of normal range. The model development cohort included 245 patients (880 digital slides), and the external test cohort had 71 patients (258 digital slides). Deep learning models were trained independently on hematoxylin and eosin, periodic acid-Schiff, periodic Schiff-methenamine silver and Masson's trichrome stained slides at multiple magnifications and integrated to predict the primary outcome of complete response to therapy at 12 months. Single-stain models showed area under the curves of 0.813, 0.841, 0.823, and 0.862, respectively. Further, integration of the four models into a multi-stain model achieved area under the curves of 0.901 and 0.840 on internal validation and external testing, respectively, which outperformed conventional clinicopathologic parameters including estimated glomerular filtration rate, chronicity index and reduction in proteinuria at three months. Decisive features uncovered by visualization uncovered for model prediction included tertiary lymphoid structures, glomerulosclerosis, interstitial fibrosis and tubular atrophy. Our study demonstrated the feasibility of utilizing deep learning on kidney pathology to predict treatment response for lupus patients. Further validation is required before the model could be implemented for risk stratification and to aid in making therapeutic decisions in clinical practice.

Combining physical model with neural networks for earthquake site response prediction

The amplitude of seismic waves will be significantly amplified near the Earth's surface, and this phenomenon is known as the seismic site response. Site response prediction is of paramount importance for the seismic-resistant building design and seismic risk assessment. However, accurately predicting site response has always been a challenge due to the incomplete physical knowledge and insufficient dataset volumes. Here, we propose an approach that combines the neural networks with classical homogeneous layered model for site response prediction. This approach exploits the potential for improving the accuracy of site response prediction from both the physical and data perspectives, which reduces the requirements for the model complexity and the training data volume. Compared to the physics-driven method, this approach reduces the estimation errors by approximately 50 % on average, and corrects the correlation between the observed and predicted results. This approach firstly reproduces the four-stage characteristics of the site response in the entire seismic band, and provides a new framework for site response prediction.

A dynamic response prediction model of nonlinear hysteretic isolators based on quasi-static characteristics and dynamic compensations

Nonlinear hysteretic isolators are widely applied in the field of vibration isolation due to their excellent energy dissipation capacity. However, the dynamic response prediction of nonlinear hysteretic isolators highly depends on accurate modeling methods for nonlinear hysteresis under varying conditions. This paper proposes a dynamic response prediction model based on quasi-static characteristics to precisely characterize the dynamic properties of nonlinear hysteretic isolators. The model employs a normalized Bouc-Wen framework, extended by polynomials, to describe the possible nonlinear stiffness and asymmetric hysteretic behaviors under quasi-static loading conditions. Additionally, global nonlinear elastic and damping terms are introduced to further capture the dynamic hysteretic characteristics over a wider range of operating velocities. The model parameters can be identified through quasi-static cyclic loading experiments and dynamic sinusoidal experiments. Then, the parametric model representing the restoring force is integrated into the dynamic equation for dynamic response prediction of the isolation object. The effectiveness of the proposed model is experimentally validated using an assembled hysteretic isolator. With limited testing data at specific amplitudes and frequencies, the resonance frequencies under steady-state excitations and the root mean square values under stochastic excitations are predicted with average accuracies of 97.40% and 97.09%, respectively.

Predicting survival, neurotoxicity and response in B-cell lymphoma patients treated with CAR-T therapy using an imaging features-based model

BackgroundThis multicentre retrospective observational study aims to develop imaging-based prognostic and predictive models for relapsed/refractory (R/R) B-cell lymphoma patients undergoing CAR-T therapy by integrating clinical data and imaging features. Specifically, our aim was to predict 3- and 6-month treatment response, overall survival (OS), progression-free survival (PFS), and the occurrence of the immune effector cell-associated neurotoxicity syndrome (ICANS).ResultsSixty-five patients of R/R B-cell lymphoma treated with CAR-T cells in two centres were included. Pre-infusion [18F]FDG PET/CT scans and clinical data were systematically collected, and imaging features, including kurtosis, entropy, maximum diameter, standardized uptake value (SUV) related features (SUVmax, SUVmean, SUVstd, SUVmedian, SUVp25, SUVp75), total metabolic tumour volume (MTVtotal), and total lesion glycolysis (TLGtotal), were extracted using the Quibim platform. The median age was 62 (range 21–76) years and the median follow-up for survivors was 10.47 (range 0.20–45.80) months. A logistic regression model accurately predicted neurotoxicity (AUC: 0.830), and Cox proportional-hazards models for CAR-T response at 3 and 6 months demonstrated high accuracy (AUC: 0.754 and 0.818, respectively). Median predicted OS after CAR-T therapy was 4.73 months for high MTVtotal and 37.55 months for low MTVtotal. Median predicted PFS was 2.73 months for high MTVtotal and 11.83 months for low MTVtotal. For all outcomes, predictive models, combining imaging features and clinical variables, showed improved accuracy compared to models using only clinical variables or imaging features alone.ConclusionThis study successfully integrates imaging features and clinical variables to predict outcomes in R/R B-cell lymphoma patients undergoing CAR-T. Notably, the identified MTVtotal cut-off effectively stratifies patients, as evidenced by significant differences in OS and PFS. Additionally, the predictive models for neurotoxicity and CAR-T response show promising accuracy. This comprehensive approach holds promise for risk stratification and personalized treatment strategies which may become a helpful tool for optimizing CAR-T outcomes in R/R lymphoma patients.

Refining antipsychotic treatment strategies in schizophrenia: discovery of genetic biomarkers for enhanced drug response prediction.

Schizophrenia (SCZ) is a severe mental disorder affecting around 1% of individuals worldwide. The variability in response to antipsychotic drugs (APDs) among SCZ patients presents a significant challenge for clinicians in determining the most effective medication. In this study, we investigated the biological markers and established a predictive model for APD response based on a large-scale genome-wide association study using 3269 Chinese schizophrenia patients. Each participant underwent an 8-week treatment regimen with one of five mono-APDs: olanzapine, risperidone, aripiprazole, quetiapine, or amisulpride. By dividing the response into ordinal groups of "high", "medium", and "low", we mitigated the bias of unclear treatment outcome and identified three novel significantly associated genetic loci in or near CDH12, WDR11, and ELAVL2. Additionally, we developed predictive models of response to each specific APDs, with accuracies ranging from 79.5% to 98.0%. In sum, we established an effective method to predict schizophrenia patients' response to APDs across three categories, integrating novel biomarkers to guide personalized medicine strategies.

DeepDR: a deep learning library for drug response prediction.

Accurate drug response prediction is critical to advancing precision medicine and drug discovery. Recent advances in deep learning (DL) have shown promise in predicting drug response; however, the lack of convenient tools to support such modeling limits their widespread application. To address this, we introduce DeepDR, the first DL library specifically developed for drug response prediction. DeepDR simplifies the process by automating drug and cell featurization, model construction, training, and inference, all achievable with brief programming. The library incorporates three types of drug features along with nine drug encoders, four types of cell features along with nine cell encoders, and two fusion modules, enabling the implementation of up to 135 DL models for drug response prediction. We also explored benchmarking performance with DeepDR, and the optimal models are available on a user-friendly visual interface. DeepDR can be installed from PyPI (https://pypi.org/project/deepdr). The source code and experimental data are available on GitHub (https://github.com/user15632/DeepDR).

Identification of key programmed cell death genes for predicting prognosis and treatment sensitivity in colorectal cancer

Colorectal cancer (CRC) ranks third in global incidence and second in mortality. However, a comprehensive predictive model for CRC prognosis, immunotherapy response, and drug sensitivity is still lacking. Various types of programmed cell death (PCD) are crucial for cancer occurrence, progression, and treatment, indicating their potential as valuable predictors. Fourteen PCD genes were collected and subjected to dimensionality reduction using regression methods to identify key hub genes. Predictive models were constructed and validated based on bulk transcriptomes and single-cell transcriptomes. Furthermore, the tumor microenvironment, immunotherapy response, and drug sensitivity profiles among patients with CRC were explored and stratified by risk. A risk score incorporating the PCD genes FABP4, AQP8, and NAT1 was developed and validated across four independent datasets. Patients with CRC who had a high-risk score exhibited a poorer prognosis. Unsupervised clustering algorithms were used to identify two molecular subtypes of CRC with distinct features. The risk score was combined with the clinical features to create a nomogram model with superior predictive performance. Additionally, patients with high-risk scores exhibited decreased immune cell infiltration, higher stromal scores, and reduced responsiveness to immunotherapy and first-line clinical drugs compared with low-risk patients. Furthermore, the top ten non-clinical first-line drugs for treating CRC were selected based on their predicted IC50 values. Our results indicate the efficacy of the model and its potential value in predicting prognosis, response to immunotherapy, and sensitivity to different drugs in patients with CRC.