FLIGHTED: Inferring Fitness Landscapes from Noisy High-Throughput Experimental Data.

Machine learning (ML) for protein design requires large protein fitness datasets generated by high-throughput experiments for training, fine-tuning, and benchmarking models. However, most models do not account for experimental noise inherent in these datasets, harming model performance and changing model rankings in benchmarking studies. Here, we develop FLIGHTED, a Bayesian method for generating fitness landscapes with calibrated errors from noisy high-throughput experimental data. We apply FLIGHTED to single-step selection assays such as phage display and to a novel high-throughput assay DHARMA that ties fitness to base editing activity. Our results show that FLIGHTED robustly generates fitness landscapes with accurate errors. We demonstrate that FLIGHTED improves model performance and enables the generation of protein fitness datasets of up to 106 variants with DHARMA. FLIGHTED can be used on any high-throughput assay and makes it easy for ML scientists to account for experimental noise when modeling protein fitness.

Use of scanning electron microscopy and energy dispersive X-ray spectroscopy to identify key fouling species during alternating tangential filtration.

Alternating tangential flow filtration (ATF) has become one of the primary methods for cell retention and clarification in perfusion bioreactors. However, membrane fouling can cause product sieving losses that limit the performance of these systems. This study used scanning electron microscopy and energy dispersive X-ray spectroscopy to identify the nature and location of foulants on 0.2 µm polyethersulfone hollow fiber membranes after use in industrial Chinese hamster ovary cell perfusion bioreactors for monoclonal antibody production. Membrane fouling was dominated by proteinaceous material, primarily host cell proteins along with some monoclonal antibody. Fouling occurred primarily on the lumen surface with much less protein trapped within the depth of the fiber. Protein deposition was also most pronounced near the inlet/exit of the hollow fibers, which are the regions with the greatest flux (and transmembrane pressure) during the cyclical operation of the ATF. These results provide important insights into the underlying phenomena governing the fouling behavior of ATF systems for continuous bioprocessing.

Neural network-derived Potts models for structure-based protein design using backbone atomic coordinates and tertiary motifs.

Designing novel proteins to perform desired functions, such as binding or catalysis, is a major goal in synthetic biology. A variety of computational approaches can aid in this task. An energy-based framework rooted in the sequence-structure statistics of tertiary motifs (TERMs) can be used for sequence design on predefined backbones. Neural network models that use backbone coordinate-derived features provide another way to design new proteins. In this work, we combine the two methods to make neural structure-based models more suitable for protein design. Specifically, we supplement backbone-coordinate features with TERM-derived data, as inputs, and we generate energy functions as outputs. We present two architectures that generate Potts models over the sequence space: TERMinator, which uses both TERM-based and coordinate-based information, and COORDinator, which uses only coordinate-based information. Using these two models, we demonstrate that TERMs can be utilized to improve native sequence recovery performance of neural models. Furthermore, we demonstrate that sequences designed by TERMinator are predicted to fold to their target structures by AlphaFold. Finally, we show that both TERMinator and COORDinator learn notions of energetics, and these methods can be fine-tuned on experimental data to improve predictions. Our results suggest that using TERM-based and coordinate-based features together may be beneficial for protein design and that structure-based neural models that produce Potts energy tables have utility for flexible applications in protein science.

The Effect of Debiasing Protein-Ligand Binding Data on Generalization.

The structured nature of chemical data means machine-learning models trained to predict protein-ligand binding risk overfitting the data, impairing their ability to generalize and make accurate predictions for novel candidate ligands. Data debiasing algorithms, which systematically partition the data to reduce bias and provide a more accurate metric of model performance, have the potential to address this issue. When models are trained using debiased data splits, the reward for simply memorizing the training data is reduced, suggesting that the ability of the model to make accurate predictions for novel candidate ligands will improve. To test this hypothesis, we use distance-based data splits to measure how well a model can generalize. We first confirm that models perform better for randomly split held-out sets than for distant held-out sets. We then debias the data and find, surprisingly, that debiasing typically reduces the ability of models to make accurate predictions for distant held-out test sets and that model performance measured after debiasing is not representative of the ability of a model to generalize. These results suggest that debiasing reduces the information available to a model, impairing its ability to generalize.

Reproducing Quantum Probability Distributions at the Speed of Classical Dynamics: A New Approach for Developing Force-Field Functors.

Modeling nuclear quantum effects is required for accurate molecular dynamics (MD) simulations of molecules. The community has paid special attention to water and other biomolecules that show hydrogen bonding. Standard methods of modeling nuclear quantum effects like Ring Polymer Molecular Dynamics (RPMD) are computationally costlier than running classical trajectories. A force-field functor (FFF) is an alternative method that computes an effective force field that replicates quantum properties of the original force field. In this work, we propose an efficient method of computing FFF using the Wigner-Kirkwood expansion. As a test case, we calculate a range of thermodynamic properties of Neon, obtaining the same level of accuracy as RPMD, but with the shorter runtime of classical simulations. By modifying existing MD programs, the proposed method could be used in the future to increase the efficiency and accuracy of MD simulations involving water and proteins.

Skeleton of Euplectella sp.: structural hierarchy from the nanoscale to the macroscale.

Structural materials in nature exhibit remarkable designs with building blocks, often hierarchically arranged from the nanometer to the macroscopic length scales. We report on the structural properties of biosilica observed in the hexactinellid sponge Euplectella sp. Consolidated, nanometer-scaled silica spheres are arranged in well-defined microscopic concentric rings glued together by organic matrix to form laminated spicules. The assembly of these spicules into bundles, effected by the laminated silica-based cement, results in the formation of a macroscopic cylindrical square-lattice cagelike structure reinforced by diagonal ridges. The ensuing design overcomes the brittleness of its constituent material, glass, and shows outstanding mechanical rigidity and stability. The mechanical benefits of each of seven identified hierarchical levels and their comparison with common mechanical engineering strategies are discussed.

Biological glass fibers: correlation between optical and structural properties.

Biological systems have, through the course of time, evolved unique solutions for complex optical problems. These solutions are often achieved through a sophisticated control of fine structural features. Here we present a detailed study of the optical properties of basalia spicules from the glass sponge Euplectella aspergillum and reconcile them with structural characteristics. We show these biosilica fibers to have a distinctive layered design with specific compositional variations in the glass/organic composite and a corresponding nonuniform refractive index profile with a high-index core and a low-index cladding. The spicules can function as single-mode, few-mode, or multimode fibers, with spines serving as illumination points along the spicule shaft. The presence of a lens-like structure at the end of the fiber increases its light-collecting efficiency. Although free-space coupling experiments emphasize the similarity of these spicules to commercial optical fibers, the absence of any birefringence, the presence of technologically inaccessible dopants in the fibers, and their improved mechanical properties highlight the advantages of the low-temperature synthesis used by biology to construct these remarkable structures.

Elastomeric transistor stamps: reversible probing of charge transport in organic crystals.

We introduce a method to fabricate high-performance field-effect transistors on the surface of freestanding organic single crystals. The transistors are constructed by laminating a monolithic elastomeric transistor stamp against the surface of a crystal. This method, which eliminates exposure of the fragile organic surface to the hazards of conventional processing, enables fabrication of rubrene transistors with charge carrier mobilities as high as approximately 15 cm2/V.s and subthreshold slopes as low as 2nF.V/decade.cm2. Multiple relamination of the transistor stamp against the same crystal does not affect the transistor characteristics; we exploit this reversibility to reveal anisotropic charge transport at the basal plane of rubrene.