Daily Summary for 2026-10-21

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: It's the twenty-first of October, twenty twenty-six, and this is the day's research.

Marcus: One new paper came out today.

Ines: I'm Ines.

Marcus: And I'm Marcus, here with Yuki.

Ines: We'll take the day in one pass, then pull out the paper we're staying with.

Marcus: Start us off with Data-driven multi-agent modelling of calcium interactions in cell culture.

The summary: Ines: Welcome to the twenty first of October, twenty twenty six. Today we focus on better understanding calcium signaling dynamics in cell culture using data driven modeling because clearer pictures allow us to observe systems with greater detail.

Marcus: We are comparing two methods for system identification and parameter discovery for governing ordinary differential equations.

Yuki: Specifically the Constrained Regularized Least-Squares Method CRLSM and Physics Informed Neural Networks PINN.

Ines: This comparison is in the work titled Data driven multi agent modelling of calcium interactions in cell culture PINN vs Regularized Least-squares.

Marcus: What does this comparison mean for our understanding?

Yuki: It helps us characterize biological systems with greater detail by providing clearer pictures of these processes.

Ines: So the focus is on how we are trying to better understand calcium signaling dynamics in cell culture using data driven modeling.

Marcus: The methods being compared are CRLSM and PINN for discovering parameters in ordinary differential equations.

Yuki: The study uses these methods to see if one provides a clearer picture of the underlying system behavior.

Ines: We are comparing those two approaches in the paper Data driven multi agent modelling of calcium interactions in cell culture PINN vs Regularized Least-squares.

Marcus: Does this comparison illuminate how we observe and characterize biological systems with greater detail?

Yuki: Yes getting a clearer picture of these processes allows us to see biological systems with greater detail.

Ines: PINNs failed to match CRLSM performance today.

Marcus: They did not provide fair parameter estimation under their setup.

Yuki: This suggests room for improvement through tuning and uncertainty quantification.

Ines: PINNs are promising but need more careful tuning for future research.

Marcus: Agreed, uncertainty quantification is key now to see real gains.

Yuki: Data-driven multi-agent modelling of calcium interactions in cell culture: PINN vs Regularized Least-squares is worth hearing because it compares these methods for modeling calcium delivery.

Ines: That comparison helps us understand the trade-offs clearly.

Marcus: It allows better observation and characterization of processes like calcium signaling in cell culture.

Yuki: Data-driven discovery of dynamics in biological systems is valuable for observing such complex processes.

Lucky paper: 2505.20327: Ines: Welcome back to our discussion on Data driven multi agent modelling of calcium interactions in cell culture PINN vs Regularized Least-squares.

Marcus: We're moving into the background for a moment, looking at the core ideas behind this work from Aurora Poggia et al.

Yuki: It starts by saying that fundamental cell mechanisms like excitation contraction and gene expressions all result from the Ca2+ intracellular signal.

Ines: Observing and controlling this flow is critical when we look at situations like hypertension, heart disease, and diabetes.

Marcus: They mention that modeling biological systems usually involves very large entities communicating with each other, which leads to a large state subject to the curse of dimensionality.

Yuki: The Multi Agent System or MAS was first introduced for identifying biological behaviors and chemical reactions with limited state dimensions.

Ines: So they propose a methodology for characterization by comparing the Constrained Regularized Least-Squares Method CRLSM and Physics Informed Neural Networks PINN for system identification.

Marcus: The consensus problem they use aims to find the velocity of the agent that leads to a final consensus in the population where all agents reach some agreement.

Yuki: That means translating into a system having similar states after a certain period of time across all agents.

Ines: They develop a framework for microscopy guided machine learning modelling of the dynamic organization of living cells.

Marcus: The calcium concentration in each cell is modeled by a simple integrator and parameters are tuned using this reduced order dataset from microscopy images.

Yuki: For system identification methods, they compare CRLSM, which minimizes the regularized least-squares minimization problem, against PINN.

Ines: The PINN loss function combines a physical loss associated with equation twenty-one and a data fitting loss associated with equation twenty-two.

Marcus: In their numerical experiments, they simulate a single cell first and then generalize it to other cells assuming the behavior of the remaining cells follows what they observed in the data.

Yuki: They analyze this by fitting parameters using LSM which minimizes over the sum of squared residuals P p(rp)two where rp is yp minus y hat p.

Ines: For a group model, CRLSM performs under the assumption that each cell is characterized by a different feed gamma i.

Marcus: The analysis splits cells into three groups based on how well the model captures their behavior.

Yuki: The CRLSM shows that for setup I, cell forty-three has an MSE of zero point zero zero one four.

Ines: Meanwhile, the PINN training process shows that the learned parameter k is notably near to the initial condition which they suggest is probably due to the PINN's training procedure concurrently minimizing both losses.

Marcus: But they note that it seems the model is limited by the initial state for the remaining cells which prevents additional model adaptation.

Yuki: The conclusion states that CRLSM achieves reliable parameter estimation and sufficient accuracy compared to PINN, which did not achieve sufficient accuracy compared to CRLSM.

Ines: The resulting CRLSM is able to approximate well suited parameters to describe the system which biologists can utilize to detect anomalies in the cell culture.

Marcus: Future research should focus on improving the neural network architecture of PINN and enforcing initial condition limitations and introducing Graph Neural Networks GNNs combined with a Neural Operator framework.