Abstract: this article presents a novel approach for automated refractive index matching in polyacrylamide gels used in capillary electrophoresis CE with the aim of optimizing separation resolution. Traditional methods require iterative manual adjustments that are time-consuming and observer-dependent. GelOpt is a system that employs a deep learning model trained with a broad dataset of CE separations to predict optimal refractive index adjustment conditions according to sample characteristics and the desired separation profile. Results show a reduction in optimization time by a factor of 10, an improvement in separation resolution close to 15 and greater reproducibility, accelerating workflows in drug discovery and proteomic analysis.
Introduction: capillary electrophoresis CE is a high-sensitivity technique for separating charged molecules according to their electrophoretic mobility. Resolution largely depends on the precise match between the refractive index of the polymeric gel and that of the buffer. Mismatches produce peak broadening and loss of efficiency. Conventional methods for adjusting the refractive index are experimental and demand numerous runs and operational expertise. This work proposes automating that adjustment through an artificial intelligence system capable of predicting and controlling gel composition in real time.
System architecture: GelOpt integrates three main modules: input processing, deep learning prediction and optimization control. The processing module analyzes the sample using multimodal techniques to extract a feature vector that feeds the neural network. The prediction block uses a hybrid CNN RNN architecture that extracts features and models temporal dependencies of separation profiles. The controller translates the prediction into adjustments in a microfluidic system that mixes acrylamide and bisacrylamide and uses a polarization modulation interferometry PMI sensor to monitor the refractive index during polymerization.
Input processing: sample characterization incorporates UV Vis spectroscopy to determine concentrations and detect absorbing components that affect the refractive index, mass spectrometry to identify compositions and m over z ratios, and a prior knowledge database PKDB with refractive index values for common analytes. The input vector X includes absorbance values, absorbance derivatives, m over z of detected components and RI values extracted from PKDB when available.
Deep learning prediction: the CNN network acts as a feature extractor from X, producing a compact representation C. An RNN processes C to predict the target refractive index and optionally a series of sequential adjustments if polymerization requires corrections during gel hardening. The output y corresponds to the target RI that optimizes expected resolution in CE.
Control and feedback: the system implements an adaptive control loop that adjusts monomer concentration according to the difference between measured RI and target RI. The adopted discrete control law is Ri plus 1 equals Ri plus K times y minus Ri where Ri is the current refractive index at step i, y is the predicted target and K is a feedback gain adjusted through adaptive learning based on PMI sensor dynamics and polymerization speed.
Experimental design: the training base was obtained through a fractional factorial design 2 to the 7 minus 1 covering 128 RI combinations using standard acrylamide and bisacrylamide ratios. Data augmentation techniques were applied to expand available variability. Validation included standard protein separations (BSA, myoglobin, carbonic anhydrase), analysis of low molecular weight drug mixtures, reproducibility tests with several technicians and robustness assays against temperature and pressure variations.
Metrics and analysis: resolution was quantified using the de Garis separation factor and reproducibility using the coefficient of variation between operators. Regression analysis was performed to correlate gel composition parameters and experimental conditions with observed resolution. The PMI sensor demonstrated precision of 0.01 RI units under tested conditions.
Results: GelOpt achieved an average resolution improvement of 15 compared to the manual protocol and reduced optimization time from approximately 8 hours to 45 minutes in typical tasks, representing a process acceleration by a factor of 10. Inter-operator variability dropped to a coefficient of variation close to 5. Robustness tests indicated a degradation in RI consistency of around 0.03 units per degree of temperature when adequate thermal regulation was not applied, suggesting incorporating thermal control for industrial environments.
Discussion: the main advantage of GelOpt is the automation of a traditionally manual and expert-dependent task, while also enabling exploitation of complex relationships between sample composition and optimal gel conditions that are not evident through heuristics. Limitations mainly derive from training set coverage: samples with completely novel compounds or extreme temperature ranges will require data expansion and possibly on-site continuous learning strategies.
Technical contributions: the combination of CNN for spectral pattern extraction and RNN for separation sequence modeling proves especially suitable for this problem. The real-time adaptive control loop supported by PMI sensor allows dynamic corrections during polymerization, reducing experimental oversampling and improving reproducibility. The use of a fractional experimental design and data augmentation techniques enabled training robust models with reasonable experimental cost.
Practical applications: GelOpt has direct impact in pharmaceutical industries and proteomic research where CE speed and reproducibility are critical. Use examples include purity characterization of pharmacological candidates, analysis of complex mixtures in clinical proteomics and quality control in reagent manufacturing. Integration with automated workflows and cloud analysis services facilitates its adoption at scale.
Operational aspects and deployment: for industrial implementation it is recommended to integrate GelOpt with cloud platforms for data storage and continuous model training, leveraging aws and azure cloud services for scalability and security. The architecture supports model delivery as a service and integration with business intelligence and visualization tools such as power bi for monitoring and reporting.
Improvement recommendations: expand the PKDB database with experimental values under non-standard conditions, incorporate additional sensors to measure humidity and pressure, and develop online learning modules to adapt K and the predictive model to new laboratory conditions. It is also advisable to create automated validation pipelines to ensure regulatory compliance in pharmaceutical environments.
Impact and outlook: automation of refractive index adjustment with GelOpt reduces operational costs, accelerates development cycles and improves quality of separation data. In the medium term, the technology can be integrated with AI agents for laboratory orchestration and with artificial intelligence solutions for companies demanding real-time analysis and continuous improvements in analytical processes.
About Q2BSTUDIO: Q2BSTUDIO is a software and custom application development company specialized in artificial intelligence solutions, cybersecurity and aws and azure cloud services. We offer custom software and custom applications that integrate artificial intelligence models, AI agents and business intelligence tools. Our services include data pipeline implementation, business intelligence services consulting, power bi integration and secure cloud deployments. Q2BSTUDIO supports clients in improving analytical and scientific processes through personalized AI solutions for companies with focus on reproducibility, security and scalability.
Keywords and positioning: custom applications, custom software, artificial intelligence, cybersecurity, aws and azure cloud services, business intelligence services, AI for companies, AI agents, power bi.
Conclusions: GelOpt demonstrates that the combination of deep learning, microfluidics and high-precision sensors can automate and optimize refractive index matching in gels for CE, improving resolution, reducing times and increasing reproducibility. Industrial adoption requires expanding robustness against environmental variables and establishing data pipelines that enable continuous learning. Q2BSTUDIO can collaborate in the development and integration of these solutions, providing expertise in custom software, artificial intelligence, cybersecurity and cloud services to bring GelOpt from the laboratory to the production environment.
Contact and next steps: for more information about integrating GelOpt into workflows and custom services, contact Q2BSTUDIO to evaluate requirements, define project scope and design a scalable and secure proof of concept.
