Predicting Cancer Drug Response In Vivo by Learning an Optimal Feature Selection of Tumour Molecular Profiles

Investigating the effects of artificial intelligence on the personalization of breast cancer management: a systematic study

BACKGROUND

Providing appropriate specialized treatment to the right patient at the right time is considered necessary in cancer management. Targeted therapy tailored to the genetic changes of each breast cancer patient is a desirable feature of precision oncology, which can not only reduce disease progression but also potentially increase patient survival. The use of artificial intelligence alongside precision oncology can help physicians by identifying and selecting more effective treatment factors for patients.

METHOD

A systematic review was conducted using the PubMed, Embase, Scopus, and Web of Science databases in September 2023. We performed the search strategy with keywords, namely: Breast Cancer, Artificial intelligence, and precision Oncology along with their synonyms in the article titles. Descriptive, qualitative, review, and non-English studies were excluded. The quality assessment of the articles and evaluation of bias were determined based on the SJR journal and JBI indices, as well as the PRISMA2020 guideline.

RESULTS

Forty-six studies were selected that focused on personalized breast cancer management using artificial intelligence models. Seventeen studies using various deep learning methods achieved a satisfactory outcome in predicting treatment response and prognosis, contributing to personalized breast cancer management. Two studies utilizing neural networks and clustering provided acceptable indicators for predicting patient survival and categorizing breast tumors. One study employed transfer learning to predict treatment response. Twenty-six studies utilizing machine-learning methods demonstrated that these techniques can improve breast cancer classification, screening, diagnosis, and prognosis. The most frequent modeling techniques used were NB, SVM, RF, XGBoost, and Reinforcement Learning. The average area under the curve (AUC) for the models was 0.91. Moreover, the average values for accuracy, sensitivity, specificity, and precision were reported to be in the range of 90-96% for the models.

CONCLUSION

Artificial intelligence has proven to be effective in assisting physicians and researchers in managing breast cancer treatment by uncovering hidden patterns in complex omics and genetic data. Intelligent processing of omics data through protein and gene pattern classification and the utilization of deep neural patterns has the potential to significantly transform the field of complex disease management.

An innovative prognostic auxiliary for colon adenocarcinoma based on zinc finger protein genes

Background

The carcinogenesis and progression of colon adenocarcinoma (COAD) are intensively related to the abnormal expression of the zinc finger (ZNF) protein genes. We aimed to employ these genes to provide a reliable prognosis and treatment stratification tool for COAD patients.

Methods

Cox and the least absolute shrinkage and selection operator (LASSO) regression analysis were applied, utilizing The Cancer Genome Atlas (TCGA) metadata, to build a ZNF protein gene-based prognostic model. Using this model, patients in the training cohort and testing cohort (GSE17537) were labelled as either high or low risk. Kaplan-Meier (KM) survival analysis and time-dependent receiver operating characteristic (ROC) curve analysis were performed in the patients with opposite risk status to assess the predictive ability in each cohort. The potentiality of the mechanism was explored by the estimation of stromal and immune cells in malignant tumor tissues using expression data (ESTIMATE), single-sample gene set enrichment analysis (ssGSEA), gene set enrichment analysis (GSEA), Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG). Finally, the degrees of expression of model genes were validated by immunohistochemistry (IHC).

Results

The prognostic model consisting of INSM1, PHF21B, RNF138, SYTL4, WRNIP1, ZNF585B, and ZNF514, classified patients into opposite risk statuses. Patients in the high-risk subset had a considerably lower chance of surviving compared to those in the low-risk subset. There is a high probability that these model genes were attached to immune-related biological processes, which can be confirmed by the results of the above mechanistic methods. Moreover, patients in the low-risk subset also significantly outperformed the patients in the high-risk subset when calculating immune cells and function scores. Drug sensitivity and tumor immune dysfunction and exclusion (TIDE) analyses showed a clear difference in the immunological and chemotherapeutic efficacy predictions within the two risk groups. Additionally, the degrees of expression of model genes in high-risk and low-risk subsets presented great discrepancies.

Conclusions

The signature may be applied as a predictive classifier to shepherd special medication for COAD patients.

Large-Scale Machine Learning Analysis Reveals DNA Methylation and Gene Expression Response Signatures for Gemcitabine-Treated Pancreatic Cancer

Background: Gemcitabine is a first-line chemotherapy for pancreatic adenocarcinoma (PAAD), but many PAAD patients do not respond to gemcitabine-containing treatments. Being able to predict such nonresponders would hence permit the undelayed administration of more promising treatments while sparing gemcitabine life-threatening side effects for those patients. Unfortunately, the few predictors of PAAD patient response to this drug are weak, none of them exploiting yet the power of machine learning (ML). Methods: Here, we applied ML to predict the response of PAAD patients to gemcitabine from the molecular profiles of their tumors. More concretely, we collected diverse molecular profiles of PAAD patient tumors along with the corresponding clinical data (gemcitabine responses and clinical features) from the Genomic Data Commons resource. From systematically combining 8 tumor profiles with 16 classification algorithms, each of the resulting 128 ML models was evaluated by multiple 10-fold cross-validations. Results: Only 7 of these 128 models were predictive, which underlines the importance of carrying out such a large-scale analysis to avoid missing the most predictive models. These were here random forest using 4 selected mRNAs [0.44 Matthews correlation coefficient (MCC), 0.785 receiver operating characteristic-area under the curve (ROC-AUC)] and XGBoost combining 12 DNA methylation probes (0.32 MCC, 0.697 ROC-AUC). By contrast, the hENT1 marker obtained much worse random-level performance (practically 0 MCC, 0.5 ROC-AUC). Despite not being trained to predict prognosis (overall and progression-free survival), these ML models were also able to anticipate this patient outcome. Conclusions: We release these promising ML models so that they can be evaluated prospectively on other gemcitabine-treated PAAD patients.

Patient-derived tumor models in cancer research: Evaluation of the oncostatic effects of melatonin

The development of new anticancer therapies tends to be very slow. Although their impact on potential candidates is confirmed in preclinical studies, ∼95 % of these new therapies are not approved when tested in clinical trials. One of the main reasons for this is the lack of accurate preclinical models. In this context, there are different patient-derived models, which have emerged as a powerful oncological tool: patient-derived xenografts (PDXs), patient-derived organoids (PDOs), and patient-derived cells (PDCs). Although all these models are widely applied, PDXs, which are created by engraftment of patient tumor tissues into mice, is considered more reliable. In fundamental research, the PDX model is used to evaluate drug-sensitive markers and, in clinical practice, to select a personalized therapeutic strategy. Melatonin is of particular importance in the development of innovative cancer treatments due to its oncostatic impact and lack of adverse effects. However, the literature regarding the oncostatic effect of melatonin in patient-derived tumor models is scant. This review aims to describe the important role of patient-derived models in the development of anticancer treatments, focusing, in particular, on PDX models, as well as their use in cancer research. This review also summarizes the existing literature on the anti-tumoral effect of melatonin in patient-derived models in order to propose future anti-neoplastic clinical applications.

New Therapeutic Perspectives in Prostate Cancer: Patient-Derived Organoids and Patient-Derived Xenograft Models in Precision Medicine

One of the major goals in the advancement of basic cancer research focuses on the development of new anticancer therapies. To understand the molecular mechanisms of cancer progression, acquired drug resistance, and the metastatic process, the use of preclinical in vitro models that faithfully summarize the properties of the tumor in patients is still a necessity. The tumor is represented by a diverse group of cell clones, and in recent years, to reproduce in vitro preclinical tumor models, monolayer cell cultures have been supplanted by patient-derived xenograft (PDX) models and cultured organoids derived from the patient (PDO). These models have proved indispensable for the study of the tumor microenvironment (TME) and its interaction with tumor cells. Prostate cancer (PCa) is the most common neoplasia in men in the world. It is characterized by genomic instability and resistance to conventional therapies. Despite recent advances in diagnosis and treatment, PCa remains a leading cause of cancer death. Here, we review the studies of the last 10 years as the number of papers is growing very fast in the field. We also discuss the discovered limitations and the new challenges in using the organoid culture system and in using PDXs in studying the prostate cancer phenotype, performing drug testing, and developing anticancer molecular therapies.

Radiation Sensitivity: The Rise of Predictive Patient-Derived Cancer Models

Patient-derived cancer models have been used for decades to improve our understanding of cancer and test anticancer treatments. Advances in radiation delivery have made these models more attractive for studying radiation sensitizers and understanding an individual patient's radiation sensitivity. Advances in the use of patient-derived cancer models lead to a more clinically relevant outcome, although many questions remain regarding the optimal use of patient-derived xenografts and patient-derived spheroid cultures. The use of patient-derived cancer models as personalized predictive avatars through mouse and zebrafish models is discussed, and the advantages and disadvantages of patient-derived spheroids are reviewed. In addition, the use of large repositories of patient-derived models to develop predictive algorithms to guide treatment selection is discussed. Finally, we review methods for establishing patient-derived models and identify key factors that influence their use as both avatars and models of cancer biology.

Ten-year distant-recurrence risk prediction in breast cancer by CanAssist Breast (CAB) in Dutch sub-cohort of the randomized TEAM trial

BACKGROUND

Hormone receptor (HR)-positive, HER2/neu-negative breast cancers have a sustained risk of recurrence up to 20 years from diagnosis. TEAM (Tamoxifen, Exemestane Adjuvant Multinational) is a large, multi-country, phase III trial that randomized 9776 women for the use of hormonal therapy. Of these 2754 were Dutch patients. The current study aims for the first time to correlate the ten-year clinical outcomes with predictions by CanAssist Breast (CAB)-a prognostic test developed in South East Asia, on a Dutch sub-cohort that participated in the TEAM. The total Dutch TEAM cohort and the current Dutch sub-cohort were almost similar with respect to patient age and tumor anatomical features.

METHODS

Of the 2754 patients from the Netherlands, which are part of the original TEAM trial, 592 patients' samples were available with Leiden University Medical Center (LUMC). The risk stratification of CAB was correlated with outcomes of patients using logistic regression approaches entailing Kaplan-Meier survival curves, univariate and multivariate cox-regression hazards model. We used hazard ratios (HRs), the cumulative incidence of distant metastasis/death due to breast cancer (DM), and distant recurrence-free interval (DRFi) for assessment.

RESULTS

Out of 433 patients finally included, the majority, 68.4% had lymph node-positive disease, while only a minority received chemotherapy (20.8%) in addition to endocrine therapy. CAB stratified 67.5% of the total cohort as low-risk [DM = 11.5% (95% CI, 7.6-15.2)] and 32.5% as high-risk [DM = 30.2% (95% CI, 21.9-37.6)] with an HR of 2.90 (95% CI, 1.75-4.80; P < 0.001) at ten years. CAB risk score was an independent prognostic factor in the consideration of clinical parameters in multivariate analysis. At ten years, CAB high-risk had the worst DRFi of 69.8%, CAB low-risk in the exemestane monotherapy arm had the best DRFi of 92.7% [vs CAB high-risk, HR, 0.21 (95% CI, 0.11-0.43), P < 0.001], and CAB low-risk in the sequential arm had a DRFi of 84.2% [vs CAB high-risk, HR, 0.48 (95% CI, 0.28-0.82), P = 0.009].

CONCLUSIONS

Cost-effective CAB is a statistically robust prognostic and predictive tool for ten-year DM for postmenopausal women with HR+/HER2-, early breast cancer. CAB low-risk patients who received exemestane monotherapy had an excellent ten-year DRFi.

Data augmentation and multimodal learning for predicting drug response in patient-derived xenografts from gene expressions and histology images

Patient-derived xenografts (PDXs) are an appealing platform for preclinical drug studies. A primary challenge in modeling drug response prediction (DRP) with PDXs and neural networks (NNs) is the limited number of drug response samples. We investigate multimodal neural network (MM-Net) and data augmentation for DRP in PDXs. The MM-Net learns to predict response using drug descriptors, gene expressions (GE), and histology whole-slide images (WSIs). We explore whether combining WSIs with GE improves predictions as compared with models that use GE alone. We propose two data augmentation methods which allow us training multimodal and unimodal NNs without changing architectures with a single larger dataset: 1) combine single-drug and drug-pair treatments by homogenizing drug representations, and 2) augment drug-pairs which doubles the sample size of all drug-pair samples. Unimodal NNs which use GE are compared to assess the contribution of data augmentation. The NN that uses the original and the augmented drug-pair treatments as well as single-drug treatments outperforms NNs that ignore either the augmented drug-pairs or the single-drug treatments. In assessing the multimodal learning based on the MCC metric, MM-Net outperforms all the baselines. Our results show that data augmentation and integration of histology images with GE can improve prediction performance of drug response in PDXs.

A Boolean-based machine learning framework identifies predictive biomarkers of HSP90-targeted therapy response in prostate cancer

Precision medicine has emerged as an important paradigm in oncology, driven by the significant heterogeneity of individual patients' tumour. A key prerequisite for effective implementation of precision oncology is the development of companion biomarkers that can predict response to anti-cancer therapies and guide patient selection for clinical trials and/or treatment. However, reliable predictive biomarkers are currently lacking for many anti-cancer therapies, hampering their clinical application. Here, we developed a novel machine learning-based framework to derive predictive multi-gene biomarker panels and associated expression signatures that accurately predict cancer drug sensitivity. We demonstrated the power of the approach by applying it to identify response biomarker panels for an Hsp90-based therapy in prostate cancer, using proteomic data profiled from prostate cancer patient-derived explants. Our approach employs a rational feature section strategy to maximise model performance, and innovatively utilizes Boolean algebra methods to derive specific expression signatures of the marker proteins. Given suitable data for model training, the approach is also applicable to other cancer drug agents in different tumour settings.

A novel epithelial-mesenchymal transition (EMT)-related gene signature of predictive value for the survival outcomes in lung adenocarcinoma

Lung adenocarcinoma (LUAD) is a remarkably heterogeneous and aggressive disease with dismal prognosis of patients. The identification of promising prognostic biomarkers might enable effective diagnosis and treatment of LUAD. Aberrant activation of epithelial-mesenchymal transition (EMT) is required for LUAD initiation, progression and metastasis. With the purpose of identifying a robust EMT-related gene signature (E-signature) to monitor the survival outcomes of LUAD patients. In The Cancer Genome Atlas (TCGA) database, least absolute shrinkage and selection operator (LASSO) analysis and cox regression analysis were conducted to acquire prognostic and EMT-related genes. A 4 EMT-related and prognostic gene signature, comprising dickkopf-like protein 1 (DKK1), lysyl oxidase-like 2 (LOXL2), matrix Gla protein (MGP) and slit guidance ligand 3 (SLIT3), was identified. By the usage of datum derived from TCGA database and Western blotting analysis, compared with adjacent tissue samples, DKK1 and LOXL2 protein expression in LUAD tissue samples were significantly higher, whereas the trend of MGP and SLIT3 expression were opposite. Concurrent with upregulation of epithelial markers and downregulation of mesenchymal markers, knockdown of DKK1 and LOXL2 impeded the migration and invasion of LUAD cells. Simultaneously, MGP and SLIT3 silencing promoted metastasis and induce EMT of LUAD cells. In the TCGA-LUAD set, receiver operating characteristic (ROC) analysis indicated that our risk model based on the identified E-signature was superior to those reported in literatures. Additionally, the E-signature carried robust prognostic significance. The validity of prediction in the E-signature was validated by the three independent datasets obtained from Gene Expression Omnibus (GEO) database. The probabilistic nomogram including the E-signature, pathological T stage and N stage was constructed and the nomogram demonstrated satisfactory discrimination and calibration. In LUAD patients, the E-signature risk score was associated with T stage, N stage, M stage and TNM stage. GSEA (gene set enrichment analysis) analysis indicated that the E-signature might be linked to the pathways including GLYCOLYSIS, MYC TARGETS, DNA REPAIR and so on. In conclusion, our study explored an innovative EMT based prognostic signature that might serve as a potential target for personalized and precision medicine.

Patient-derived cancer models: Valuable platforms for anticancer drug testing

Molecularly targeted treatments and immunotherapy are cornerstones in oncology, with demonstrated efficacy across different tumor types. Nevertheless, the overwhelming majority metastatic disease is incurable due to the onset of drug resistance. Preclinical models including genetically engineered mouse models, patient-derived xenografts and two- and three-dimensional cell cultures have emerged as a useful resource to study mechanisms of cancer progression and predict efficacy of anticancer drugs. However, variables including tumor heterogeneity and the complexities of the microenvironment can impair the faithfulness of these platforms. Here, we will discuss advantages and limitations of these preclinical models, their applicability for drug testing and in co-clinical trials and potential strategies to increase their reliability in predicting responsiveness to anticancer medications.

Interpretable Machine Learning Models to Predict the Resistance of Breast Cancer Patients to Doxorubicin from Their microRNA Profiles

Doxorubicin is a common treatment for breast cancer. However, not all patients respond to this drug, which sometimes causes life-threatening side effects. Accurately anticipating doxorubicin-resistant patients would therefore permit to spare them this risk while considering alternative treatments without delay. Stratifying patients based on molecular markers in their pretreatment tumors is a promising approach to advance toward this ambitious goal, but single-gene gene markers such as HER2 expression have not shown to be sufficiently predictive. The recent availability of matched doxorubicin-response and diverse molecular profiles across breast cancer patients permits now analysis at a much larger scale. 16 machine learning algorithms and 8 molecular profiles are systematically evaluated on the same cohort of patients. Only 2 of the 128 resulting models are substantially predictive, showing that they can be easily missed by a standard-scale analysis. The best model is classification and regression tree (CART) nonlinearly combining 4 selected miRNA isoforms to predict doxorubicin response (median Matthew correlation coefficient (MCC) and area under the curve (AUC) of 0.56 and 0.80, respectively). By contrast, HER2 expression is significantly less predictive (median MCC and AUC of 0.14 and 0.57, respectively). As the predictive accuracy of this CART model increases with larger training sets, its update with future data should result in even better accuracy.

Tissue-specific identification of multi-omics features for pan-cancer drug response prediction

Current statistical models for drug response prediction and biomarker identification fall short in leveraging the shared and unique information from various cancer tissues and multi-omics profiles. We developed mix-lasso model that introduces an additional sample group penalty term to capture tissue-specific effects of features on pan-cancer response prediction. The mix-lasso model takes into account both the similarity between drug responses (i.e., multi-task learning), and the heterogeneity between multi-omics data (multi-modal learning). When applied to large-scale pharmacogenomics dataset from Cancer Therapeutics Response Portal, mix-lasso enabled accurate drug response predictions and identification of tissue-specific predictive features in the presence of various degrees of missing data, drug-drug correlations, and high-dimensional and correlated genomic and molecular features that often hinder the use of statistical approaches in drug response modeling. Compared to tree lasso model, mix-lasso identified a smaller number of tissue-specific features, hence making the model more interpretable and stable for drug discovery applications.

•

Pan-cancer cell lines provide a test bench for exploring gene-drug relationships

•

Multi-omics data were integrated with pharmacological profiles for joint modeling

•

Mix-lasso identifies tissue-specific biomarkers predictive of multi-drug responses

•

Mix-lasso provides small number of stable features for drug discovery applications

SynergyFinder 3.0: an interactive analysis and consensus interpretation of multi-drug synergies across multiple samples

SynergyFinder (https://synergyfinder.fimm.fi) is a free web-application for interactive analysis and visualization of multi-drug combination response data. Since its first release in 2017, SynergyFinder has become a popular tool for multi-dose combination data analytics, partly because the development of its functionality and graphical interface has been driven by a diverse user community, including both chemical biologists and computational scientists. Here, we describe the latest upgrade of this community-effort, SynergyFinder release 3.0, introducing a number of novel features that support interactive multi-sample analysis of combination synergy, a novel consensus synergy score that combines multiple synergy scoring models, and an improved outlier detection functionality that eliminates false positive results, along with many other post-analysis options such as weighting of synergy by drug concentrations and distinguishing between different modes of synergy (potency and efficacy). Based on user requests, several additional improvements were also implemented, including new data visualizations and export options for multi-drug combinations. With these improvements, SynergyFinder 3.0 supports robust identification of consistent combinatorial synergies for multi-drug combinatorial discovery and clinical translation.

Predicting Objective Response Rate (ORR) in Immune Checkpoint Inhibitor (ICI) Therapies with Machine Learning (ML) by Combining Clinical and Patient-Reported Data

ICIs are a standard of care in several malignancies; however, according to overall response rate (ORR), only a subset of eligible patients benefits from ICIs. Thus, an ability to predict ORR could enable more rational use. In this study a ML-based ORR prediction model was built, with patient-reported symptom data and other clinical data as inputs, using the extreme gradient boosting technique (XGBoost). Prediction performance for unseen samples was evaluated using leave-one-out cross-validation (LOOCV), and the performance was evaluated with accuracy, AUC (area under curve), F1 score, and MCC (Matthew’s correlation coefficient). The ORR prediction model had a promising LOOCV performance with all four metrics: accuracy (75%), AUC (0.71), F1 score (0.58), and MCC (0.4). A rather good sensitivity (0.58) and high specificity (0.82) of the model were seen in the confusion matrix for all 63 LOOCV ORR predictions. The two most important symptoms for predicting the ORR were itching and fatigue. The results show that it is possible to predict ORR for patients with multiple advanced cancers undergoing ICI therapies with a ML model combining clinical, routine laboratory, and patient-reported data even with a limited size cohort.