Chris A. Kieslich

Background: Patients with multiple myeloma (MM) are at increased risk for light chain cardiac amyloidosis (CA) due to high circulating levels of light chains. ALCA is a clinical diagnosis based on biopsy, cardiac imaging findings of restrictive cardiomyopathy, and serum monoclonal proteins. To overcome these challenges, we have developed machine learning models that accurately predict ALCA status while identifying key diagnostic features based on routine laboratory measurements and using limited data. Methods: Utilizing electronic medical data (EMR) data, we built support vector machine models (SVMs) using the MISTIC framework. Data including diagnosis history, treatment, and laboratory values for patients with MM or monoclonal gammopathy of unknown significance (MGUS) were collected as a retrospective cohort study at Moffit Cancer Center and Research Institute between 2014 and 2019. To utilize the wealth of information in longitudinal EMR data, we fit regression models to the EMR data and used the parameters and characteristics of the fits as features for our SVMs. We tested three different sets of features: descriptive statistics alone, parameters and characteristics of fit only, and both the descriptive statistics and parameters of fit (DescFits). A classifier using raw data at a timepoint proximal to the date of ALCA diagnostic imaging (EchoAdj), as well as analogous random forest and XGBoost models were used for comparison. Findings: SVM classifiers trained on the DescFits feature set (AUC= 1·0) outperformed blind model predictions of the EchoAdj model (AUC = 0·36) and other conventional machine learning approaches in identifying which patients were diagnosed with ALCA. The DescFits model selected features describing the temporal relationship of physiologic parameters including hematologic state, and cardiac and renal function. Interpretation: We demonstrated that the MISTIC SVM framework and the shape of EMR data can aid in identifying patients with cardiac amyloidosis without cardiac biopsy or imaging. Funding: This work was in part supported by funding from the National Institutes of Health (NCATS KL2TROO1429 to MCF and NIGMS R35GM147164 to CAK) and Moffitt Cancer Center and Research Institute’s Junior Scientist Partnership Award (to MCF and IBR). MAC is a trainee in the Medical Scientist Training Program supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number T32GM163993.

The dynamics of anisotropic particles are dictated by forces and torques that can be challenging to mathematically represent in computer simulations. Several data-driven approaches have been developed to approximate these interactions, but they often rely on having large amounts of training data that may be practically difficult to generate. Here, we extend a framework we recently developed for approximating anisotropic pair potentials to the approximation of pairwise forces and torques. The framework uses multivariate polynomials and physics-motivated coordinate transformations to produce accurate approximations using limited amounts of data. We first derive expressions relating the force and torque to partial derivatives of the approximated potential energy with respect to the transformed coordinates used to represent the particle configuration. We then explore several options for approximating the same pairwise interactions from measurements of the forces and torques, and we critically assess their accuracy using model two- and three-dimensional shape-anisotropic nanoparticles as test cases. We find that interpolation of the pairwise potential energy produces the best result when it is known, but force and torque matching (regression) is a viable strategy when only the force and torque are available.

The properties of films assembled by drying colloidal-particle suspensions depend sensitively on both the particles and the processing conditions, making them challenging to engineer. In this work, we develop and test an inverse-design strategy based on surrogate modeling to identify conditions that yield a target film structure. We consider a two-component hard-sphere colloidal suspension whose designable parameters are the particle sizes, the initial composition of particles, and the drying rate. Film drying is simulated approximately using Brownian dynamics. Surrogate models based on Gaussian process regression (GPR) and Chebyshev polynomial interpolation are trained on a loss function, computed from the simulated film structures, that guides the design process. We find the surrogate models to be effective for both approximation and optimization using only a small number of samples of the loss function. The GPR models are typically slightly more accurate than polynomial interpolants trained using comparable amounts of data, but the polynomial interpolants are more computationally convenient. This work has important implications not only for designing colloidal materials but also more broadly as a strategy for engineering nonequilibrium assembly processes.

Enzymes are catalysts in biochemical reactions that, by definition, increase rates of reactions without being altered or destroyed. However, when that enzyme is a protease, a subclass of enzymes that hydrolyze other proteins, and that protease is in a multiprotease system, protease-as-substrate dynamics must be included, challenging assumptions of enzyme inertness, shifting kinetic predictions of that system. Protease-on-protease inactivating hydrolysis can alter predicted protease concentrations used to determine pharmaceutical dosing strategies. Cysteine cathepsins are proteases capable of cathepsin cannibalism, where one cathepsin hydrolyzes another with substrate present, and misunderstanding of these dynamics may cause miscalculations of multiple proteases working in one proteolytic network of interactions occurring in a defined compartment. Once rates for individual protease-on-protease binding and catalysis are determined, proteolytic network dynamics can be explored using computational models of cooperative/competitive degradation by multiple proteases in one system, while simultaneously incorporating substrate cleavage. During parameter optimization, it was revealed that additional distraction reactions, where inactivated proteases become competitive inhibitors to remaining, active proteases, occurred, introducing another network reaction node. Taken together, improved predictions of substrate degradation in a multiple protease network were achieved after including reaction terms of autodigestion, inactivation, cannibalism, and distraction, altering kinetic considerations from other enzymatic systems, since enzyme can be lost to proteolytic degradation. We compiled and encoded these dynamics into an online platform (https://plattlab.shinyapps.io/catKLS/) for individual users to test hypotheses of specific perturbations to multiple cathepsins, substrates, and inhibitors, and predict shifts in proteolytic network reactions and system dynamics.

Applications or “apps” for computers and mobile devices are ubiquitous in modern technology. Several apps are available for teaching chemical engineering and are disseminated inexpensively or are freely available online. This paper highlights software apps that chemical engineering instructors may use “off the shelf” for iOS, Mathematica, Excel, R, WebMO, MATLAB, and COMSOL. Brief notes on additional resources and time demand are provided for those interested in customizing or developing their own educational apps.

In this article, we present (1) a feature selection algorithm based on nonlinear support vector machine (SVM) for fault detection and diagnosis in continuous processes and (2) results for the Tennessee Eastman benchmark process. The presented feature selection algorithm is derived from the sensitivity analysis of the dual C ‐SVM objective function. This enables simultaneous modeling and feature selection paving the way for simultaneous fault detection and diagnosis, where feature ranking guides fault diagnosis. We train fault‐specific two‐class SVM models to detect faulty operations, while using the feature selection algorithm to improve the accuracy and perform the fault diagnosis. Our results show that the developed SVM models outperform the available ones in the literature both in terms of detection accuracy and latency. Moreover, it is shown that the loss of information is minimized with the use of feature selection techniques compared to feature extraction techniques such as principal component analysis (PCA). This further facilitates a more accurate interpretation of the results. © 2018 American Institute of Chemical Engineers AIChE J , 65: 992–1005, 2019

Every two years groups worldwide participate in the Critical Assessment of Protein Structure Prediction (CASP) experiment to blindly test the strengths and weaknesses of their computational methods. CASP has significantly advanced the field but many hurdles still remain, which may require new ideas and collaborations. In 2012 a web-based effort called WeFold, was initiated to promote collaboration within the CASP community and attract researchers from other fields to contribute new ideas to CASP. Members of the WeFold coopetition (cooperation and competition) participated in CASP as individual teams, but also shared components of their methods to create hybrid pipelines and actively contributed to this effort. We assert that the scale and diversity of integrative prediction pipelines could not have been achieved by any individual lab or even by any collaboration among a few partners. The models contributed by the participating groups and generated by the pipelines are publicly available at the WeFold website providing a wealth of data that remains to be tapped. Here, we analyze the results of the 2014 and 2016 pipelines showing improvements according to the CASP assessment as well as areas that require further adjustments and research.

The complement system is our first line of defense against foreign pathogens, but when it is not properly regulated, complement is implicated in the pathology of several autoimmune and inflammatory disorders. Compstatin is a peptidic complement inhibitor that acts by blocking the cleavage of complement protein C3 to the proinflammatory fragment C3a and opsonin fragment C3b. In this study, we aim to identify druglike small-molecule complement inhibitors with physicochemical, geometric, and binding properties similar to those of compstatin. We employed two approaches using various high-throughput virtual screening methods, which incorporate molecular dynamics (MD) simulations, pharmacophore model design, energy calculations, and molecular docking and scoring. We have generated a library of 274 chemical compounds with computationally predicted binding affinities for the compstatin binding site of C3. We have tested subsets of these chemical compounds experimentally for complement inhibitory activity, using hemolytic assays, and for binding affinity, using microscale thermophoresis. As a result, although none of the compounds showed inhibitory activity, compound 29 was identified to exhibit weak competitive binding against a potent compstatin analogue, therefore validating our computational approaches. Additional docking and MD simulation studies suggest that compound 29 interacts with C3 residues, which have been shown to be important in binding of compstatin to the C3c fragment of C3. Compound 29 is amenable to physicochemical optimization to acquire inhibitory properties. Additionally, it is possible that some of the untested compounds will demonstrate binding and inhibition in future experimental studies.