Algorithmic discovery, development and personalized selection of higher-order drug cocktails : A label-free live-cell imaging & secretomics approach

• An upward trend in clinical pharmacology is the use of multiple drugs to combat complex and co-occurring diseases due to better efficacy, decreased toxicity and reduced risk of evolving resistance. Despite high late-stage attrition rates and the need for multi drug treatments, most drug discovery and development efforts are still mainly focused on new one-size-fits-all monotherapies. This is unfortunate given the complex, heterogeneous and often only partially understood pathophysiology of many diseases. In this context, polypharmacotherapies hold strong potential, especially when patient tailored. However, as of today, the personalized combination therapy area remains vastly unexplored. A major reason is lack of standardized and robust tools that allow systematic in vitro drug combination sensitivity testing of different disease models and patient derived cells.This thesis fills in this lack by introducing two methodological frameworks, namely COMBImageDL and COMBSecretomics, designed to enable systematic second- and higher-order drug combination studies within and beyond cancer pharmacology. They include advanced quality control procedures, non-parametric resampling statistics to quantify uncertainty and a data driven methodology to evaluate response patterns and discern higher- from lower- and single-drug effects. Both are based on a standardized and reproducible format that could be employed with any experimental platform that provides the required raw data. COMBImageDL searches exhaustively for drug cocktails that induce changes in cell viability and time evolving cell culture morphology by employing conventional endpoint synergy analyses jointly with quantitative label-free live-cell imaging. Deep neural network learning, MapReduce parallel processing and method-specific parameter tuning are key components of the design. The purely phenotypic functionality of COMBImageDL is extended by COMBSecretomics, which searches exhaustively for drug cocktails that can modify, or even reverse malfunctioning secretomic patterns. It processes complex datasets involving drug treated cells observed before and after being stimulated by relevant proteins. Finally, the highest single agent method is generalized for higher-order drug combination analysis and adjusted for secreted protein profiles.The frameworks were used in five pharmacological studies being industrial, academic and clinical collaborations in areas where novel and personalized multi drug regimens are highly needed; oncology (acute myeloid leukemia and glioblastoma multiforme) and osteoarthritis. These studies demonstrate intriguing drug combination findings and in general the great potential of tools like COMBImageDL and COMBSecretomics to accelerate the discovery and development of novel potent polypharmacotherapeutic candidates.

Ämnesord

MEDICIN OCH HÄLSOVETENSKAP  -- Medicinsk bioteknologi -- Biomedicinsk laboratorievetenskap/teknologi (hsv//swe)

MEDICAL AND HEALTH SCIENCES  -- Medical Biotechnology -- Biomedical Laboratory Science/Technology (hsv//eng)

Nyckelord

drug combination discovery and development

personalized pharmacotherapy

higher-order drug combination analysis

quantitative label-free live-cell imaging

secretomics

cell-cell communication

deep learning

convolutional neural networks

data mining

MapReduce

resampling

generalized highest single agent

COMBImageDL

COMBSecretomics

Bioinformatics

Bioinformatik

Pharmacology

Farmakologi

Engineering Science

Teknisk fysik

Publikations- och innehållstyp

vet (ämneskategori)

dok (ämneskategori)