Dynamical model-based experiment design for drug repositioning

TL;DR

This work introduces a dynamical model–based experimental design for drug repositioning that iteratively learns system dynamics and drug effects while guiding subsequent experiments to identify effective drug combinations. By formulating the disease as a linear system with unknown $A$ and $b$ and drug influence $B$, the authors define a cost that encodes steering the state toward a healthy target $x_h$ and develop an iterative procedure to expand candidate drug sets, score drugs by their contribution, and refine the model from experimental data. In silico results show that continuous-time identification improves parameter accuracy and that near-optimal drug combinations can be discovered within realistic testing budgets, achieving small gaps (often $<10\%$) for up to six drugs. The approach highlights the potential of coupling dynamical modelling with systematic experimental design to accelerate drug repurposing, while acknowledging nonlinearities, topology uncertainty, and exploration–exploitation challenges as directions for future work.

Abstract

Computational methods in drug repositioning can help to conserve resources. In particular, methods based on biological networks are showing promise. Considering only the network topology and knowledge on drug target genes is not sufficient for quantitative predictions or predictions involving drug combinations. We propose an iterative procedure alternating between system identification and drug response experiments. Data from experiments are used to improve the model and drug effect knowledge, which is then used to select drugs for the next experiments. Using simulated data, we show that the procedure can identify nearly optimal drug combinations.

Figures (4)

• Figure 1: Pipeline of the model-based experimental cycle. The iterative modelling-experiment cycle is illustrated with red arrows.
• Figure 2: A typical topology generated during one replicate.
• Figure 3: Performance of the estimation methods across 20 replicates measured by the Frobenius norm of the error between the estimated $A$ and true $A$. Black: iterative continuous-time scheme. Gray: explicit Euler discretisation. The vertical red line indicates the drug testing capacity used in the simulation (1 high and 25 low frequency tests) in §\ref{['sec:result']}.
• Figure 4: Left: performance gap between the best found drug combinations and the true optima. Right: the costs for the best found combinations compared to the disease state cost. Both measures are computed for 1 to 6-drug combinations and the statistics across 20 replicates are presented in box plot format.