Resolving causal genes from quantitative trait loci (QTL) remains fundamentally limited by linkage disequilibrium. We developed an interpretable machine learning framework that utilizes multi-omic data, which enables it to capture higher-order nonlinear genotype-phenotype relationships and allows conditional evaluation of genetic variants, enabling statistical decorrelation of linked loci. Applied to Saccharomyces cerevisiae segregants across chemical stress conditions, our models, built by integrating genomics and phenomics data, achieved >75% prediction accuracy and identified known causal genes, including MKT1 (genotoxic stress) and IRA2 (osmotic stress). SHAP-based analysis recovered 56% of the validated pleiotropic genes, compared with 36% by conventional contingency testing. Integration with genome-scale metabolic models built using transcriptomic data revealed pathway enrichments distinguishing high-growing strains, including carbon transport, glycolysis, and oxidative phosphorylation. Notably, gene regulatory network analysis identified a novel function for PDR8 in protein mannosylation and cell wall integrity—functions extending beyond its role in drug resistance. This framework demonstrates that interpretable machine learning, coupled with systems biology, transforms QTL associations into mechanistic biological insight.

Effective monitoring and control of bioprocesses is indispensable for sustaining an optimal extracellular environment for microbial growth and metabolite production. While near-infrared (NIR) spectroscopy is extensively used for real-time monitoring, its integration into control remains scarce due to batch-to-batch and within-batch variations inherent to bioprocesses. This work presents a maintenance, monitoring, and control framework that combines online NIR spectroscopy with offline process data to address these variations. A robust re-calibration approach using historical data to handle the variations is presented. The framework is empirically validated using the Lactococcus lactis NZ9000 fermentation process, where the glucose concentration is maintained at 30 g/L for 12 h using a PID controller. The within-batch variation handling is demonstrated in-silico using experimental data from the fermentation process, providing preliminary evidence for its feasibility. It is shown that the recursively updated calibration model improves the prediction of metabolite concentrations, achieving tight glucose control (within 4% of the set point) during fermentation. Additionally, in-silico studies show the application of statistical metrics to identify drift and indicate the need for model re-calibration.

Biomass productivities in shake flasks are often not reproduced in bioreactors for plant cell cultures due to change in hydrodynamics. Considering shake flask biomass productivity as benchmark, this study employs shake flask geometries as a model system to understand hydrodynamic changes with volume and identify suitable scale-up criteria for plant cell cultivations, with minimal cost and time, given their slow growth time, using computational fluid dynamics (CFD) and experiments. Cultivation of Viola odorata cells in increasing flask volumes (100-3000 mL) revealed no significant change in biomass productivity. CFD analysis indicated that volumetric oxygen mass transfer coefficient (kLa), increased up to 1000 mL and then decreased, due to saturation of energy dissipation rates (kL is a function of energy dissipation rates) and decreasing interfacial area. The unaffected biomass concentration, despite decreased kLa, suggests that kLa may not be a significant scale-up parameter. Instead, maintaining a constant shear environment, indicated by power per unit volume saturation at higher volumes, was proposed as a suitable scale-up parameter for V. odorata cell cultivation in bioreactors. Moreover, the decrease in velocity difference between fluid layers with increased flask volume, indicated that minimizing velocity gradients in bioreactors could help achieve shake flask biomass productivity.

Cultivation of plant cell cultures in conventional bioreactors designed for microbial cells often results in decrease of biomass productivity as compared to that in shake flasks, presumably due to the imbalance between the mass transfer requirements and compromise with cell viability. Hit and trial methods for bioreactor design are generally performed to achieve high biomass productivity in the bioreactor. In this study, a rational approach has been adopted to choose a suitable impeller for Viola odorata cell suspension culture using computational fluid dynamics (CFD). A two-phase CFD model was employed to characterize the non-Newtonian fluid dynamics of the plant cell suspension in a stirred tank reactor using different impeller designs, a setric, Rushton and marine impeller. The simulations were performed adopting Euler-Euler approach for the two-phase flow and dispersed $$\:\kappa\:-\epsilon\:$$ turbulence model. The numerical model was validated with good agreement with experimental determination of volumetric mass transfer coefficient. The impact of impeller design was then investigated on critical process parameters like mixing, oxygen mass transfer and shear. The developed CFD model demonstrated that setric impeller is a suitable choice for V. odorata cell cultivation among the three impellers offering low-shear environment at equivalent velocity magnitudes at reactor bottom with higher cell-lift capabilities which is preferable in high cell-density plant cell cultivations.

Microreactors are an essential part of modular chemical systems involved in the on-demand production of chemicals such as nanomaterials, pharmaceuticals, specialty chemicals, etc. Model-based nonlinear predictive control of microreactors is a challenging task due to the high online computational cost associated with developing and maintaining high-order first-principles nonlinear models. In this work, we propose a nonlinear data-driven model predictive control (NMPC) scheme for nanoparticle production in microreactors. In this paper, a non-linear Auto Regressive Exogenous Neural Network model (NARX-NN) is developed with the flow rates of the reactants as inputs and the peak value of the absorbance spectra (an indirect measure of the average size of nanoparticles) as output by performing a set of experiments in Corning Advanced-Flow TM Reactors (AFR). Typically, producing a new desired average size nanoparticle on-demand is done by manual changes in the flow rates of reactants. In this work, a nonlinear model predictive controller using the identified NARX-NN model is formulated to track a change in the set point, the peak value of the spectra. The formulated controller with the identified NARX-NN model is demonstrated via the simulation studies. It is shown that the proposed NMPC with the NARX-NN model performs well in different scenarios of silver nanoparticle production.

Scaled-down vehicles provide a controlled and repeatable environment for testing autonomous driving algorithms, expediting development while mitigating risks associated with full-scale vehicle testing. This paper introduces a meticulously designed scaled-down electric vehicle, emphasizing vehicle state measurement, precise control, perception sensing, and electric safety. In contrast to conventional scaled platforms built on radio-controlled cars comprising actuators lacking measurements, precision, and reliability, our approach utilizes actuators with feedback, enhancing precision and torque across all speed ranges. Apart from the conventional sensors used in the existing scaled platforms, our vehicle incorporates novel sensor modules to measure wheel angular velocities, steering angles, and battery cell voltages, which can also be seamlessly integrated into these platforms. We demonstrate the vehicle's motion with manual and autonomous operation and showcase its features for validation.

This paper details the intricacies in the design of power supplies for varied voltage and current requirements in scaled-down electric vehicle platforms. It includes the development of protective circuits safeguarding electrical and electronic components from overvoltage, undervoltage, reverse voltage, inrush current, and surge current. Furthermore, the paper explores a sophisticated sensing module designed for continuous monitoring of battery and cell voltages. The circuits are implemented and tested on a one-tenth-scale electric vehicle, DEFT, operating with a 6-cell LiPo battery.

Perfect adaptation is a ubiquitous phenomenon in biology, characterized by the ability of living cells to maintain chemical levels despite disturbances. This adaptation is said to be robust when it holds even after parameter variations. This property of the system is called Robust Perfect Adaptation (RPA). When certain variables lose this property or undesired chemical species acquire RPA property, it results in disease conditions. Metabolic adaptation in extreme tumor microenvironment is a hallmark of cancer cells and allows them to evade death and develop drug resistance. Hence, it is important to investigate the robustness properties of cancer metabolism using a system-theoretical approach. In this work, we topologically investigate the robust perfect adaptation property of large-scale metabolic reaction networks of three types of cancers using the theory proposed by Hirono, Gupta, and Khammash, 2023 (https://arxiv.org/abs/2307.07444). We enumerate the RPA properties of all biologically relevant reactions in large-scale metabolic models of cancer cells, along with the corresponding healthy cells. Finally, perform a comparative analysis to identify drug targets based on the differences in the RPA properties between healthy and cancer cell metabolisms.

The semi-supervised machine learning approach is an integrated calibration-free modelling framework for identifying reaction systems from spectral data using minimal prior information and it is validated with experimental data obtained in a micro-reactor.

A priori parameter identifiability analysis is a first step towards building a reliable and robust model of biochemical reaction networks. Differential algebra-based approaches are widely used to study the a priori structural identifiability of nonlinear systems. However, these approaches fail to yield any results in complex reaction networks. Recently, Varghese et al, IFAC-paper Online, 2018, have shown that the input-output map can be computed by proposing a linear transformation of the system equations for chemical reaction networks. Subsequently, the identifiability of reaction networks can be studied using this input-output map. In this work, we investigate this input-output map for establishing the conditions for global identifiability of enzymatic reaction networks arising in systems biology. Furthermore, we derive the reparameterized form of kinetic parameters that can be uniquely identified for enzymatic kinetic models. We illustrate the results using several common enzymatic kinetic rate expressions in systems biology. Copyright (c) 2024 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

This work considers an important problem of identifying the dynamics of chemical reaction networks from time-series data. We propose an approach to identify complex chemical reaction networks (CRN) from concentration data using the concept of sparse model identification. Particularly, we demonstrate challenges associated with the application of the sparse identification of nonlinear dynamics (SINDy) and its variants to data obtained from CRNs. We develop a SINDy-CRN algorithm based on the properties of CRNs for identifying governing equations of a CRN. The proposed algorithm is illustrated using a numerical simulation example.

Model predictive control (MPC) is a popular approach for trajectory optimization in practical robotics applications. MPC policies can optimize trajectory parameters under kinodynamic and safety constraints and provide guarantees on safety, optimality, generalizability, interpretability, and explainability. However, some behaviors are complex and it is difficult to hand-craft an MPC objective function. A special class of MPC policies called Learnable-MPC addresses this difficulty using imitation learning from expert demonstrations. However, they require the demonstrator and the imitator agents to be identical which is hard to satisfy in many real world applications of robotics. In this paper, we address the practical problem of training Learnable-MPC policies when the demonstrator and the imitator do not share the same dynamics and their state spaces may have a partial overlap. We propose a novel approach that uses a generative adversarial network (GAN) to minimize the Jensen-Shannon divergence between the state-trajectory distributions of the demonstrator and the imitator. We evaluate our approach on a variety of simulated robotics tasks of DeepMind Control suite and demonstrate the efficacy of our approach at learning the demonstrator's behavior without having to copy their actions.

Reactive species (RS) play significant roles in many disease contexts. Despite their crucial roles in diseases including cancer, the RS are not adequately modeled in the genome-scale metabolic (GSM) models, which are used to understand cell metabolism in disease contexts. We have developed a scalable RS reactions module that can be integrated with any Recon 3D-derived human metabolic model, or after fine-tuning, with any metabolic model. With RS-integration, the GSM models of three cancers (basal-like triple negative breast cancer (TNBC), high grade serous ovarian carcinoma (HGSOC) and colorectal cancer (CRC)) built from Recon 3D, precisely highlighted the increases/decreases in fluxes (dysregulation) occurring in important pathways of these cancers. These dysregulations were not prominent in the standard cancer models without the RS module. Further, the results from these RS-integrated cancer GSM models suggest the following decreasing order in the ease of ferroptosis-targeting to treat the cancers: TNBC > HGSOC > CRC.

In reaction systems, state estimators are used to improve the quality of estimates using measurements and process models with the number of moles or concentrations as states. The model of reaction systems can be reformulated in the extent domain with the reaction and flow extents as states. This work exploits the properties of the extents, such as nonnegativity and monotonicity, and formulates nonlinear and linear Shape-Constrained Moving Horizon Estimators (SCMHE) for reaction systems. It is shown that the linear SCMHE is a quadratic programming problem, and hence, it is computationally less expensive. The performance of the SCMHE schemes is compared with the Extended Kalman filters (EKF) and MHE in the concentration domain via simulation studies using two examples, namely, gas-phase isothermal batch reactor and lactic acid production in a fed-batch reactor. It is shown that the linear SCMHE provides better performance than EKF with a similar average computational time, while nonlinear SCMHE is computationally cheaper and performs better than the MHE in the concentration or mole domain.

In this work, we formulate a safe human-in-loop reinforcement problem by allowing human experts to specify safe actions during the training. We propose two algorithms, safe Q-learning and partially safe Q-learning, which use constrained action spaces during training to find a safe optimal policy. In the case of partially safe Q-learning, the concept of safety ratio is introduced to provide the agent an ability to explore while safety is guaranteed. The proposed algorithms are corroborated by performing simulation studies for four different environments of varying complexity. It is shown that a partially safe Q-learning approach outperforms safe Q-learning and Q-learning on various tasks.

Elucidating genotype-phenotype or variant-to-function relationships remains a challenge in quantitative genetics. For quantitative traits, causal SNPs act either additively or epistatically, resulting in complex interactions that are difficult to dissect molecularly. Here, we developed gene co-expression networks and genome-scale metabolic models for all combinations of causal SNPs of yeast sporulation efficiency, a quantitative trait, to determine how genetic interactions drive phenotypic variation. Analysis of expression networks identified SNP-SNP interaction-dependent changes in the connectivity of crucial metabolic regulators. Using an efficient thresholding and model extraction algorithm, Localgini+iMAT, we integrated the gene expression data of all SNP combinations in the yeast genome-scale metabolic model, resulting in 16 SNP-specific metabolic models. Genome-scale differential flux analysis was used to identify differentially activated metabolic reactions across several metabolic pathways in each SNP-specific model. This analysis revealed causal reactions in six major metabolic pathways, explaining the differences in observed sporulation efficiency. Finally, the differential regulation of the pentose phosphate pathway in specific SNP combinations suggested autophagy as a pentose phosphate pathway-dependent compensatory mechanism for increasing sporulation efficiency. Our study highlights how genetic polymorphisms in transcription factors can interact to modulate metabolic pathways in a well-studied complex trait in yeast, providing insights into genome-wide association studies (GWAS) variants of metabolic traits.### Competing Interest StatementThe authors have declared no competing interest.

ABSTRACT Elucidating genotype-phenotype or variant-to-function relationships remains a pivotal challenge in genetics. For quantitative traits, causal SNPs act either additively or epistatically, resulting in complex interactions which are difficult to dissect molecularly. Here, we developed gene co-expression networks and genome-scale metabolic models for all combinations of causal SNPs of yeast sporulation efficiency, a quantitative trait, to determine how genetic interactions drive phenotypic variation. Analysis of expression networks identified SNP-SNP interaction-dependent changes in the connectivity of crucial metabolic regulators. These regulators revealed how some SNP combinations added to higher phenotypic value. Genome-scale differential flux analysis was used to identify differentially activated metabolic reactions across multiple metabolic pathways in each SNP-specific model. This analysis identified causal reactions explaining the observed sporulation efficiency differences. Finally, the differential regulation of the pentose phosphate pathway in specific SNP combinations suggested autophagy as a pentose phosphate pathway-dependent compensatory mechanism for increasing sporulation efficiency. Our study presents a modelling framework that has the potential to unravel specific causal metabolic pathways and reactions regulated by SNPs and their combinations, thus providing insights into GWAS variants of metabolic diseases.

In this paper, we propose an end-to-end autonomous driving architecture for safe maneuvering in heterogeneous traffic using a reinforcement learning (RL) algorithm. Using the proposed architecture we develop an RL agent that can make driving decisions directly from the sensor data. We formulate the autonomous driving problem as a Markov Decision Process and propose different architectures using Deep $Q -$Networks for two types of sensor data - top view images of the autonomous vehicle (AV) and its surrounding vehicles and information on relative position and velocities of the surrounding vehicles w.r.t the AV. We consider a highway scenario and analyze the performance of the RL agent using the proposed architectures using the highway-env simulator. We compare the driving performance of the AV for both sensor types and discuss their efficacy under varying traffic densities.

Accurate position estimation, especially in indoor environments, presents a formidable challenge despite significant advancements in estimation methodologies. The integration of additional sensors is often imperative to refine estimates, particularly within indoor settings. The choice of additional sensors depends on the specific applications and operational range. This study focuses on devising a position estimation framework utilizing the internal sensors such as the Inertial Measurement Unit (IMU) and wheel encoders within a scaled-down Electric Vehicle. The proposed approach entails a combination of crafting a digital filter and implementing a Kalman filter to effectively attenuate noise and errors in the IMU data, thereby enhancing the position estimate, and the methodology is validated through experiments. This work contributes to the broader discourse on advancing indoor position estimation while bypassing the need for external positioning systems.