Search

Despite improvements, there are still two main reasons why clinical trials fail. These are drug ineffectiveness and drug-induced toxicity, which are primarily the result of poor cohort selection and patient monitoring. Machine learningis an area of artificial intelligence that allows computers to learn without being explicitly programmed by analysing and drawing conclusions from data patterns. This thesis investigated innovative strategies for modernising the process of drug clinical development by incorporating machine learning-based algorithms to uncover clinically significant patterns from various sources of data, culminating in classification models. This thesis work focused into three primary research issues. The first research issue concerns the use of an machine learning approach to identify known and novel predictors of dose limiting toxicity by analysing clinical, baseline blood biochemistry (i.e., prior to starting the phase I), and genetic data derived from a previously conducted phase Ib clinical trial in metastatic colorectal cancer patients treated with FOLFIRI (folinic acid, 5-fluorouracil, irinotecan) plus bevacizumab regimen. The analyses pipeline used includes a step selecting the best predictors based on importance rankings; the optimal subset was then used to train models. The performance of five machine learning classification models was evaluated in order to select the best classifier. The Random Forest model performed best during cross-validation, with a mean Matthews correlation coefficient of 0.549 and a mean accuracy of 80.4%; at baseline, the top predictors of dose-limiting toxicity were haemoglobin, serum glutamic oxaloacetic transaminase (SGOT), and albumin. The second thesis question aims to evaluate the relationship between genetic variations covering over 60 candidate genes and carboplatin, taxane, and bevacizumab-induced toxicities in patients with ovarian cancer enrolled in a phase IV study. Machine learning techniques were employed to investigate and prioritise germline genetic variants associated with drug-induced toxicities, specifically hypertension, hemalogical toxicity, non hemalogical toxicity and proteinuria. The Boruta algorithm was used in a cross-validation fashion to determine the significance of SNPs for predicting toxicities. The process revealed which SNPs were actually important, and those were subsequently used to train each XGBoost classifier. During cross-validation toxicities models achieved reliable performances with an Matthews correlation coefficient score that varied from 0.375 to 0.410 (Accuracy from 0.696 to 0.789). The third study topic aims to develop and validate a predictive machine learning model capable of classifying hepatocellular carcinoma patients based on their cancer progression status six months after treatment using their DNA methylation profile. The genome-wide DNA methylation profile of 374 primary tumor specimens was used in combination of Machine Learning algorithms (Recursive Features Selection, Boruta) to capture early tumor progression features. The subsets of probes obtained were used to train and validate Random Forest models to predict a Progression Free Survival greater or less than 6 months. A model based on 34 epigenetic probes showed the best performance, scoring 0.80 Accuracy and 0.51 Matthews Correlation Coefficient on testset. In conclusion, this thesis present practical machine learning applications that lead to the creation of novel ways for modernising the medication clinical development process in clinical trials. The models and the evidences generated from these applications might be employed in the CT ecosystem to identify patients who are most likely to benefit from the treatment, making trials safer and faster while also cutting failure rates. Moreover, the analytic frameworks proposed in this thesis are generalizable and adaptable to outcomes and pathologies that fall far outside the sphere of pharmacology. Despite improvements, there are still two main reasons why clinical trials fail. These are drug ineffectiveness and drug-induced toxicity, which are primarily the result of poor cohort selection and patient monitoring. Machine learningis an area of artificial intelligence that allows computers to learn without being explicitly programmed by analysing and drawing conclusions from data patterns. This thesis investigated innovative strategies for modernising the process of drug clinical development by incorporating machine learning-based algorithms to uncover clinically significant patterns from various sources of data, culminating in classification models. This thesis work focused into three primary research issues. The first research issue concerns the use of an machine learning approach to identify known and novel predictors of dose limiting toxicity by analysing clinical, baseline blood biochemistry (i.e., prior to starting the phase I), and genetic data derived from a previously conducted phase Ib clinical trial in metastatic colorectal cancer patients treated with FOLFIRI (folinic acid, 5-fluorouracil, irinotecan) plus bevacizumab regimen. The analyses pipeline used includes a step selecting the best predictors based on importance rankings; the optimal subset was then used to train models. The performance of five machine learning classification models was evaluated in order to select the best classifier. The Random Forest model performed best during cross-validation, with a mean Matthews correlation coefficient of 0.549 and a mean accuracy of 80.4%; at baseline, the top predictors of dose-limiting toxicity were haemoglobin, serum glutamic oxaloacetic transaminase (SGOT), and albumin. The second thesis question aims to evaluate the relationship between genetic variations covering over 60 candidate genes and carboplatin, taxane, and bevacizumab-induced toxicities in patients with ovarian cancer enrolled in a phase IV study. Machine learning techniques were employed to investigate and prioritise germline genetic variants associated with drug-induced toxicities, specifically hypertension, hemalogical toxicity, non hemalogical toxicity and proteinuria. The Boruta algorithm was used in a cross-validation fashion to determine the significance of SNPs for predicting toxicities. The process revealed which SNPs were actually important, and those were subsequently used to train each XGBoost classifier. During cross-validation toxicities models achieved reliable performances with an Matthews correlation coefficient score that varied from 0.375 to 0.410 (Accuracy from 0.696 to 0.789). The third study topic aims to develop and validate a predictive machine learning model capable of classifying hepatocellular carcinoma patients based on their cancer progression status six months after treatment using their DNA methylation profile. The genome-wide DNA methylation profile of 374 primary tumor specimens was used in combination of Machine Learning algorithms (Recursive Features Selection, Boruta) to capture early tumor progression features. The subsets of probes obtained were used to train and validate Random Forest models to predict a Progression Free Survival greater or less than 6 months. A model based on 34 epigenetic probes showed the best performance, scoring 0.80 Accuracy and 0.51 Matthews Correlation Coefficient on testset. In conclusion, this thesis present practical machine learning applications that lead to the creation of novel ways for modernising the medication clinical development process in clinical trials. The models and the evidences generated from these applications might be employed in the CT ecosystem to identify patients who are most likely to benefit from the treatment, making trials safer and faster while also cutting failure rates. Moreover, the analytic frameworks proposed in this thesis are generalizable and adaptable to outcomes and pathologies that fall far outside the sphere of pharmacology.

Cinnamic acids are an important class of phenolic compounds, which have many beneficial effects on human health but are also interesting synthetic intermediates thanks to the presence of several reactive sites. While studying the reactivity of cinnamic acids with diazonium salts from aromatic amines, an unexpected reactivity has been discovered, leading to the formation of 1,2-diaza-1,3-dienes instead of traditional diazo-coupling products. The new compounds have been fully characterized by mono and bidimensional NMR spectroscopy and mass spectrometry. Preliminary studies on the biological activity of the compounds have been carried out testing both their antibacterial and antitumor activity, leading to promising results.

Il monitoraggio terapeutico dei farmaci (TDM) è la pratica clinica di misurazione di un farmaco specifico a intervalli di tempo definiti per mantenere le concentrazioni plasmatiche all'interno di una finestra terapeutica mirata, per massimizzare l'efficacia e ridurre al minimo la tossicità. Per rendere fattibile questo approccio sono necessari metodi analitici robusti, sensibili e riproducibili per quantificare i farmaci antitumorali nei fluidi biologici. Questo progetto di dottorato si è concentrato su quattro farmaci antitumorali orali utilizzati per il trattamento di pazienti con HCC avanzato: sorafenib (SORA), regorafenib (REGO), lenvatinib (LENVA) e idarubicina (IDA), la cui sicurezza ed efficacia è oggetto di indagine come opzione terapeutica di terza linea. SORA, REGO e LENVA sono stati inoltre coinvolti in uno studio di cross-validazione (CRO-2018-83) in corso presso il Centro di Riferimento Oncologico di Aviano, il cui obiettivo primario era la cross-validazione tra il metodo LC-MS/MS di quantificazione basato sul plasma e quello basato sul Dried Blood Spot (DBS) per dimostrare che il DBS può essere utilizzato come metodo di campionamento alternativo rispetto al campione plasmatico per la quantificazione di questi farmaci. Il DBS può migliorare l'applicabilità del TDM grazie alle sue procedure di raccolta del campione meno invasiva per il paziente. Invece, la quantificazione di IDA e del suo metabolita attivo nel plasma umano era richiesta come obiettivo secondario di uno studio clinico di fase II (CRO-2017-42) in corso presso il CRO. Per tutti questi farmaci, i dati di esposizione-risposta e -tossicità sono ancora limitati o mancanti, evidenziando così la necessità di approfondire questo tipo di indagine. Per valutare la concentrazione plasmatica del farmaco nelle matrici biologiche sono necessari metodi analitici affidabili. In questo contesto, sono stati sviluppati e validati due metodi LC-MS/MS secondo linee guida EMA ed FDA per la quantificazione contemporanea di SORA, REGO e 3 metaboliti attivi (SORA-N-ossido, REGO-N-ossido e N-desmetil-REGO-N-ossido), sia in plasma umano sia in DBS. Questi metodi hanno lo stesso range analitico (50-8000 ng/mL) per SORA e REGO e (30-4000 ng/mL) per i metaboliti. Dopo le validazioni, questi metodi sono stati applicati per quantificare 66 campioni di plasma e 63 DBS ottenuti da 16 pazienti, trattati con SORA o REGO e arruolati nello studio CRO 2018 83. Tale analisi ha permesso di ottenere dati preliminari riguardanti le concentrazioni dei farmaci nei pazienti e la correlazione tra campioni accoppiati di plasma e DBS. L'applicazione di adeguate analisi statistiche per lo studio di cross-validazione ha mostrato l'assenza di una forte correlazione tra le concentrazioni plasmatiche e quelle in DBS ed ha suggerito di aumentare il numero di pazienti e di rivalutare l'effetto dell'ematocrito e del volume dello spot per migliorare tale conversione. Sono stati sviluppati altri due metodi LC-MS/MS per la quantificazione di LENVA sia nel plasma umano sia nel DBS utilizzando due diverse carte da filtro (Whatman 31 ET CHR e Whatman 903). Questi metodi sono stati validati secondo le linee guida di riferimento di EMA, FDA e per il DBS. Il metodo basato sul plasma ha un ampio range analitico (0,5-2000 ng/mL) ed è stato applicato per quantificare 24 campioni di plasma ottenuti da 6 pazienti trattati con LENVA e arruolati nello studio CRO-2018-83. Il corrispondente metodo basato su DBS ha un range analitico leggermente ridotto (5-2000 ng/mL) ed è stato applicato per l'analisi di 4 campioni di pazienti DBS. Non sono stati ancora condotti studi di correlazione a causa della scarsità di campioni dei pazienti raccolti fino ad ora. L'applicazione di questi metodi analitici agli studi in corso presso il CRO di Aviano consentirebbe di raccogliere dati utili per approfondire le conoscenze sulla possibile correlazione tra i livelli plasmatici e l'esito del trattamento o tossicità Therapeutic Drug Monitoring (TDM) is the clinical practice of measuring a specific drug at defined intervals of time to maintain plasma concentrations within a targeted therapeutic window, to maximize efficacy and minimize toxicity. Robust, sensitive, and reproducible analytical methods to quantify anticancer drugs in biological fluids are needed to make this approach feasible. This PhD project focused on four different oral anticancer drugs used to treat patients with advanced HCC: sorafenib (SORA), regorafenib (REGO), lenvatinib (LENVA), and idarubicin (IDA), whose safety and efficacy as a third-line therapy option is currently under investigation. SORA, REGO, and LENVA were also involved in a cross-validation study (CRO-2018-83) ongoing at Centro di Riferimento Oncologico di Aviano (CRO), where the primary aim was the cross-validation between the plasma-based LC-MS/MS method and the Dried Blood Spot (DBS)-based LC-MS/MS assay to demonstrate that DBS can be used alternatively to plasma sample to quantify these drugs. DBS may improve TDM applicability thanks to its more patient-friendly procedures for sample collection. Instead, the quantification of IDA and its active metabolite (idarubicinol) in human plasma was required as a secondary aim of a phase II clinical study (CRO-2017-42) ongoing at CRO. For all these drugs, exposure-response and -toxicity data are still limited or lacking, thus highlighting the necessity of deepening this type of investigation. To assess drug plasma concentration in biological matrices, reliable analytical methods are needed. In this contest, two LC-MS/MS methods were developed and validated according to international guidelines to simultaneously quantify SORA, REGO, and 3 active metabolites (SORA N oxide, REGO N oxide, and N desmethyl REGO N oxide), both in human plasma and DBS. These methods have the same analytical range (50-8000 ng/mL) for SORA and REGO and (30-4000 ng/mL) for the metabolites. After the validations, these methods were applied to quantify 66 plasma samples and 63 DBS obtained from 16 patients, treated with SORA or REGO and enrolled in CRO 2018 83. This analysis allowed to obtain preliminary data regarding the concentrations of the drugs in patients and the correlation between plasma and DBS paired samples the application of proper statistical analyses for the cross-validation study showed the absence of a strong correlation between plasma and DBS concentrations and it suggested to enlarge patients number and to re-evaluate Hct and spot-volume effect to enhance the DBS-to-plasma conversion performance. Two additional LC-MS/MS methods were developed for the quantification of LENVA both in human plasma and DBS using two different filter papers (Whatman 31 ET CHR and Whatman 903). These methods were validated according to EMA, FDA, and DBS references guidelines. The plasma-based method has a wide analytical range (0.5-2000 ng/mL) and it was applied to quantify 24 plasma samples obtained from 6 patients treated with LENVA and enrolled in CRO-2018-83. The corresponding DBS-based method has a slightly reduced analytical range (5-2000 ng/mL) and it was applied for the analysis of 4 DBS patients’ samples. No correlation studies have yet been performed due to the paucity of patients’ samples collected till now. The application of these analytical methods to the ongoing studies at the C.R.O. di Aviano should allow to collect useful data to deepen the knowledge about the possibility of establishing a correlation between drug exposure levels and patients’ outcome or toxicity development. The aim is to implement the application of TDM for these anticancer drugs in clinical practice.

Green coffee beans are particularly rich in chlorogenic acids (CGAs), and their identification and quantification are usually performed by HPLC, coupled with mass spectrometry (LC-MS). Although there are a few examples of molecularly imprinted polymers (MIPs) for chlorogenic acid (5-CQA) recognition present in the literature, none of them are based on optical fluorescence, which is very interesting given its great sensitivity. In the present manuscript, fluorescent polymeric imprinted nanoparticles were synthetized following the non-covalent approach using hydrogenated 5-O-caffeoylquinic acid (H-5-CQA) as the template. The capability of the polymer to bind 5-CQA was evaluated by HPLC and fluorescence. A real sample of coffee extract was also analyzed to verify the selectivity of the polymer. Polymer fMIP01, containing 4-vinylpyridine and a naphtalimide derivative as monomers, showed a good response to the fluorescence quenching in the range 39 μM–80 mM. In the real sample, fMIP01 was able to selectively bind 5-CQA, while caffeine was not recognized. To demonstrate this, there is a promising system that can be exploited in the design of an optical sensor for 5-CQA detection. Polymer fMIP01 was immobilized by physical entrapment on a functionalized glass surface, showing a quenching of fluorescence with an increase of the CGA concentration between 156 μM and 40 mM.