Machine Learning Models for Predicting Molecular UV-Vis Spectra with Quantum Mechanical Properties.

Accurate understanding of ultraviolet-visible (UV-vis) spectra is critical for the high-throughput synthesis of compounds for drug discovery. Experimentally determining UV-vis spectra can become expensive when dealing with a large quantity of novel compounds. This provides us an opportunity to drive computational advances in molecular property predictions using quantum mechanics and machine learning methods. In this work, we use both quantum mechanically (QM) predicted and experimentally measured UV-vis spectra as input to devise four different machine learning architectures, UVvis-SchNet, UVvis-DTNN, UVvis-Transformer, and UVvis-MPNN, and assess the performance of each method. We find that the UVvis-MPNN model outperforms the other models when using optimized 3D coordinates and QM predicted spectra as input features. This model has the highest performance for predicting UV-vis spectra with a training RMSE of 0.06 and validation RMSE of 0.08. Most importantly, our model can be used for the challenging task of predicting differences in the UV-vis spectral signatures of regioisomers.

UV-adVISor: Attention-Based Recurrent Neural Networks to Predict UV-Vis Spectra.

Ultraviolet-visible (UV-Vis) absorption spectra are routinely collected as part of high-performance liquid chromatography (HPLC) analysis systems and can be used to identify chemical reaction products by comparison to the reference spectra. Here, we present UV-adVISor as a new computational tool for predicting the UV-Vis spectra from a molecule's structure alone. UV-Vis prediction was approached as a sequence-to-sequence problem. We utilized Long-Short Term Memory and attention-based neural networks with Extended Connectivity Fingerprint Diameter 6 or molecule SMILES to generate predictive models for the UV spectra. We have produced two spectrum datasets (dataset I, N = 949, and dataset II, N = 2222) using different compound collections and spectrum acquisition methods to train, validate, and test our models. We evaluated the prediction accuracy of the complete spectra by the correspondence of wavelengths of absorbance maxima and with a series of statistical measures (the best test set median model parameters are in parentheses for model II), including RMSE (0.064), R2 (0.71), and dynamic time warping (DTW, 0.194) of the entire spectrum curve. Scrambling molecule structures with the experimental spectra during training resulted in a degraded R2, confirming the utility of the approaches for prediction. UV-adVISor is able to provide fast and accurate predictions for libraries of compounds.

Glycan specificity of neuraminidases determined in microarray format.

Neuraminidases hydrolytically remove sialic acids from glycoconjugates. Neuraminidases are produced by both humans and their pathogens, and function in normal physiology and in pathological events. Identification of neuraminidase substrates is needed to reveal their mechanism of action, but high-throughput methods to determine glycan specificity of neuraminidases are limited. Here we use two glycan labeling reactions to monitor neuraminidase activity toward glycan substrates. While both periodate oxidation and aniline-catalyzed oxime ligation (PAL) and galactose oxidase and aniline-catalyzed oxime ligation (GAL) can be used to monitor neuraminidase activity toward glycans in microtiter plates, only GAL accurately measured neuraminidase activity toward glycans displayed on a commercial glass slide microarray. Using GAL, we confirm known linkage specificities of three pneumococcal neuraminidases and obtain new information about underlying glycan specificity.

NMR characterization of an oligonucleotide model of the miR-21 pre-element.

We have used NMR spectroscopy to characterize an oligonucleotide stem loop structure based on the pre-element of an oncogenic microRNA, miR-21. This predicted stem-loop structure is cleaved from the precursor of miR-21 (pre-miR-21) by the nuclease Dicer. It is also a critical feature recognized by the protein complex that converts the primary transcript (pri-miR-21) into the pre-miRNA. The secondary structure of the native sequence is poorly defined by NMR due to rapid exchange of imino protons with solvent; however, replacement of two adjacent putative G•U base pairs with G•C base pairs retains the conformation of the hairpin observed by chemical probing and stabilizes it sufficiently to observe most of the imino proton resonances of the molecule. The observed resonances are consistent with the predicted secondary structure. In addition, a peak due to a loop uridine suggests an interaction between it and a bulged uridine in the stem. Assignment of non-exchangeable proton resonances and characterization of NOEs and coupling constants allows inference of the following features of the structure: extrahelicity of a bulged adenosine, deviation from A-form geometry in a base-paired stem, and consecutive stacking of the adenosines in the 5' side of the loop, the guanosine of the closing base pair, and a cross-strand adenosine. Modeling of the structure by restrained molecular dynamics suggests a basis for the interaction between the loop uridine, the bulged uridine in the stem, and an A•U base pair in the stem.

Association of a peptoid ligand with the apical loop of pri-miR-21 inhibits cleavage by Drosha.

We have found a small molecule that specifically inhibits cleavage of a precursor to the oncogenic miRNA, miR-21, by the microprocessor complex of Drosha and DGCR8. We identified novel ligands for the apical loop of this precursor from a screen of 14,024 N-substituted oligoglycines (peptoids) in a microarray format. Eight distinct compounds with specific affinity were obtained, three having affinities for the targeted loop in the low micromolar range and greater than 15-fold discrimination against a closely related hairpin. One of these compounds completely inhibits microprocessor cleavage of a miR-21 primary transcript at concentrations at which cleavage of another miRNA primary transcript, pri-miR-16, is little affected. The apical loop of pri-miR-21, placed in the context of pri-miR-16, is sufficient for inhibition of microprocessor cleavage by the peptoid. This compound also inhibits cleavage of pri-miR-21 containing the pri-miR-16 apical loop, suggesting an additional site of association within pri-miR-21. The reported peptoid is the first example of a small molecule that inhibits microprocessor cleavage by binding to the apical loop of a pri-miRNA.

Cyclization of peptoids by formation of boronate esters.

Introduction of conformational constraints into peptoids (N-substituted oligoglycines) will enable new applications in molecular recognition and self-assembly. Peptoids that contain both a phenylboronic acid side chain and a vicinal diol cyclize by intramolecular condensation to form boronate esters. A fluorescent indicator of free boronic acid was used to assay esterification. A galactose moiety 2 to 5 monomer units away from a boronic acid side chain in a peptoid reacts with the boronic acid in competition with the indicator. The intramolecular reaction predominates in each case, with 80-90% of the peptoid cyclized. When the diol is a simple 2,3-dihydroxypropyl group, esterification is less favored but still appreciable.

Inhibited cell spreading on polystyrene nanopillars fabricated by nanoimprinting and in situ elongation.

Polymer nanopillars (40-80 nm in diameter and 100 nm in pitch) were fabricated at high density over large areas directly on bulk tissue culture polystyrene plates using nanoimprint lithography. Nanoporous Si molds for imprinting were generated by transfer from an anodic alumina membrane. Ultrahigh aspect ratio polymer nanopillars were formed in a novel procedure using controlled elongation of the imprinted pillars during mold release. The resulting nanopillar arrays show significant changes in surface wettability upon brief O(2) plasma treatment. Human dermal fibroblasts were cultured on the nanopillar surfaces in order to study cell-substrate interaction at the nanoscale. The nanopillar topography shows strong effects on the cell morphology, with pillars of widely varying aspect ratios and surface energies resisting cell spreading. This effect on cell behavior can be rationalized in terms of the cells' requirement to form micron-scale focal adhesions. The study indicates that at the nanoscale, physical factors can supersede the effects of chemical factors on the cell-substratum interaction.

Discovering ligands for a microRNA precursor with peptoid microarrays.

We have screened peptoid microarrays to identify specific ligands for the RNA hairpin precursor of miR-21, a microRNA involved in cancer and heart disease. Microarrays were printed by spotting a library of 7680 N-substituted oligoglycines (peptoids) onto glass slides. Two compounds on the array specifically bind RNA having the sequence and predicted secondary structure of the miR-21 precursor hairpin and have specific affinity for the target in solution. Their binding induces a conformational change around the hairpin loop, and the most specific compound recognizes the loop sequence and a bulged uridine in the proximal duplex. Functional groups contributing affinity and specificity were identified, and by varying a critical methylpyridine group, a compound with a dissociation constant of 1.9 microM for the miR-21 precursor hairpin and a 20-fold discrimination against a closely-related hairpin was created. This work describes a systematic approach to discovery of ligands for specific pre-defined novel RNA structures. It demonstrates discovery of new ligands for an RNA for which no specific lead compounds were previously known by screening a microarray of small molecules.

Correlation of anisotropic cell behaviors with topographic aspect ratio.

In this study, we have used nanoimprinting to create a range of micro- and nanoscale gratings, or their combination, in bulk polystyrene plates to investigate anisotropic cell behaviors of human dermal fibroblasts with respect to the aspect ratio (depth/width) of gratings. The depth and width of the polystyrene gratings both show strong effects individually on cell alignment and elongation that are qualitatively similar to the results of other studies. However, consistent quantitative comparison of these individual parameters with different studies is complicated by the diversity of combinations of width and depth that have been tested. Instead, the aspect ratio of the gratings as a unified description of grating topography is a more consistent parameter to interpret topographic dependence of cell morphology. Both cell alignment and elongation increase with increasing aspect ratio, and even a shallow grating (aspect ratio of approximately 0.05) is sufficient to induce 80% cell alignment. Re-plotting data recently published by other groups vs. aspect ratio shows a similar dependence, despite differences in cell types and surface structures. This consistency indicates that aspect ratio is a general factor to characterize cell behaviors. The relationship of cell elongation and alignment with topographic aspect ratio is interpreted in terms of the theory of contact guidance. This model provides simplicity and flexibility in geometry design for devices and materials that interface with cells.

Manipulating cell migration and proliferation with a light-activated polypeptide.

Remote control of cells: A polypeptide has been made that stimulates proliferation and migration of cells upon photochemical activation. This light-activated polypeptide enables spatially defined control of cell populations at the scale of tissue organization; this is accomplished without physically contacting the cells or modifying their substrate. Polypeptide growth and differentiation factors modulate a wide variety of cell behaviors and can be used to manipulate cells in vitro for tissue engineering and basic studies of cell biology. To emulate in vitro the spatial aspect of growth factor function, new methods are needed to generate defined spatial gradients of activity. Polypeptide factors that are engineered to be activated with light provide a method for creating concentration gradients with the fine precision in space and time with which light can be directed. As a first test of this approach, we have chemically synthesized a polypeptide with the sequence of epidermal growth factor in which a critical glutamate is "caged" with a photoremovable group. Photolysis of this polypeptide afforded maximal mitogenic and chemokinetic activity at concentrations at which the caged factor was inactive. Spatially resolved photolysis of the factor resulted in spatial patterning of fibroblasts. This system will be useful for ex vivo tissue engineering and for investigating the interactions of cells with their matrix and the role of chemical gradients in biological pattern formation.

Organism identification using a genome sequence-independent universal microarray probe set.

There has been increasing interest and efforts devoted to developing biosensor technologies for identifying pathogens, particularly in the biothreat area. In this study, a universal set of short 12- and 13-mer oligonucleotide probes was derived independently of a priori genomic sequence information and used to generate unique species-dependent genomic hybridization signatures. The probe set sequences were algorithmically generated to be maximally distant in sequence space and not dependent on the sequence of any particular genome. The probe set is universally applicable because it is unbiased and independent of hybridization predictions based upon simplified assumptions regarding probe-target duplex formation from linear sequence analysis. Tests were conducted on microarrays containing 14,283 unique probes synthesized using an in situ light-directed synthesis methodology. The genomic DNA hybridization intensity patterns reproducibly differentiated various organisms (Bacillus subtilis, Yersinia pestis, Streptococcus pneumonia, Bacillus anthracis, and Homo sapiens), including the correct identification of a blinded "unknown" sample. Applications of this method include not only pathological and forensic genome identification in medicine and basic science, but also potentially a novel method for the discovery of unknown targets and associations inherent in dynamic nucleic acid populations such as represented by differential gene expression.

Photolithographic synthesis of peptoids.

We describe a novel photolithographic approach to the synthesis of peptoids (oligo-N-substituted glycines). This strategy enables the construction of a spatially addressable peptoid microarray, thus providing a potentially powerful tool for the discovery of protein ligands.

Patterning adhesion of mammalian cells with visible light, tris(bipyridyl)ruthenium(II) chloride, and a digital micromirror array.

Patterns of cellular adhesion were created on a surface using novel photochemistry that is stimulated with visible light. A glass surface coated with polyethylene glycol is nonadhesive to a variety of adherent mammalian cell types. Treatment of that surface with a mixture of tris(bipyridyl)ruthenium(II) chloride, ammonium persulfate, and a tryptophan derivative or tryptophan-bearing peptide in conjunction with irradiation with visible light (447 nm) made the surface adhesive to several cell types including mouse fibroblasts, human myoblasts, and human lung tumor cells. Immunostaining data suggest that tryptophan-containing peptides are crosslinked intact to the surface by this chemistry, which enables patterning of peptides containing only naturally occurring amino acids. Microscopic patterns of cellular adhesion were created with this chemistry by projecting microscopic patterns of visible light with a digital micromirror array. Using this method, regions of cellular adhesion were patterned with single-cell resolution.

Prioritized selection of oligodeoxyribonucleotide probes for efficient hybridization to RNA transcripts.

Only a small fraction of short oligonucleotide probes bind efficiently to complementary segments in long RNA transcripts. Technologies such as array-based transcript profiling and antisense control of gene expression would benefit greatly from a method for predicting probes that bind well to a given target RNA. To develop an algorithm for prioritizing selection of probes, we have analyzed predicted thermodynamic parameters for the binding of several large sets of probes to complementary RNA transcripts. The binding of five of these sets of probes to their RNA targets has been reported by others. In addition, we have used a method for light-directed synthesis of oligonucleotide arrays that we developed to generate two new arrays of surface-bound probes and measured the binding of these probes to their RNA targets. We considered predicted free energies for intramolecular base pairing of the oligonucleotide and its RNA target as well as the predicted free energy of intermolecular hybridization of probe and target. We find that a reliable predictor of probes that will hybridize significantly with their targeted transcripts is the predicted free energy of hybridization minus the predicted free energy for intramolecular folding of the probe.

A scalable high-throughput chemical synthesizer.

A machine that employs a novel reagent delivery technique for biomolecular synthesis has been developed. This machine separates the addressing of individual synthesis sites from the actual process of reagent delivery by using masks placed over the sites. Because of this separation, this machine is both cost-effective and scalable, and thus the time required to synthesize 384 or 1536 unique biomolecules is very nearly the same. Importantly, the mask design allows scaling of the number of synthesis sites without the addition of new valving. Physical and biological comparisons between DNA made on a commercially available synthesizer and this unit show that it produces DNA of similar quality.

Parallel assessment of CpG methylation by two-color hybridization with oligonucleotide arrays.

We have developed a method for the parallel analysis of multiple CpG sites in genomic DNA for their state of methylation. Hypermethylation of CpG islands within the promoters and 5' exons of genes has been found to be a mechanism of transcriptional inactivation associated with a variety of tumors. The method that we developed relies on the differential reactivity of methylated and unmethylated cytosines with sodium bisulfite, which exclusively converts unmethylated cytosines to deoxyuracils. The resulting sequence changes are determined with single-nucleotide resolution by hybridization to an oligonucleotide array. Cohybridization with a reference sample containing a different label provides an internal standard for assessment of methylation state. This method provides advantages in parallelism over existing methods of methylation analysis. We have demonstrated this technique with a region from the promoter of the tumor suppressor gene p16, which is hypermethylated in many cancers.