OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization.

AlphaFold2 revolutionized structural biology with the ability to predict protein structures with exceptionally high accuracy. Its implementation, however, lacks the code and data required to train new models. These are necessary to (1) tackle new tasks, like protein-ligand complex structure prediction, (2) investigate the process by which the model learns and (3) assess the model's capacity to generalize to unseen regions of fold space. Here we report OpenFold, a fast, memory efficient and trainable implementation of AlphaFold2. We train OpenFold from scratch, matching the accuracy of AlphaFold2. Having established parity, we find that OpenFold is remarkably robust at generalizing even when the size and diversity of its training set is deliberately limited, including near-complete elisions of classes of secondary structure elements. By analyzing intermediate structures produced during training, we also gain insights into the hierarchical manner in which OpenFold learns to fold. In sum, our studies demonstrate the power and utility of OpenFold, which we believe will prove to be a crucial resource for the protein modeling community.

The Plausibility of Sampling as an Algorithmic Theory of Sentence Processing.

Words that are more surprising given context take longer to process. However, no incremental parsing algorithm has been shown to directly predict this phenomenon. In this work, we focus on a class of algorithms whose runtime does naturally scale in surprisal-those that involve repeatedly sampling from the prior. Our first contribution is to show that simple examples of such algorithms predict runtime to increase superlinearly with surprisal, and also predict variance in runtime to increase. These two predictions stand in contrast with literature on surprisal theory (Hale, 2001; Levy, 2008a) which assumes that the expected processing cost increases linearly with surprisal, and makes no prediction about variance. In the second part of this paper, we conduct an empirical study of the relationship between surprisal and reading time, using a collection of modern language models to estimate surprisal. We find that with better language models, reading time increases superlinearly in surprisal, and also that variance increases. These results are consistent with the predictions of sampling-based algorithms.

Identification, Heterologous Expression, and Characterization of the Tolypodiol Biosynthetic Gene Cluster through an Integrated Approach.

Cyanobacteria are tremendous producers of biologically active natural products, including the potent anti-inflammatory compound tolypodiol. However, linking biosynthetic gene clusters with compound production in cyanobacteria has lagged behind that in other bacterial genera. Tolypodiol is a meroterpenoid originally isolated from the cyanobacterium HT-58-2. Here we describe the identification of the tolypodiol biosynthetic gene cluster through heterologous expression in Anabaena and in vitro protein assays of a methyltransferase found in the tolypodiol biosynthetic gene cluster. We have also identified similar biosynthetic gene clusters in cyanobacterial and actinobacterial genomes, suggesting that meroterpenoids with structural similarity to the tolypodiols may be synthesized by other microbes. We also report the identification of two new analogs of tolypodiol that we have identified in both the original and heterologous producer. This work further illustrates the usefulness of Anabaena as a heterologous expression host for cyanobacterial compounds and how integrated approaches can help to link natural product compounds with their producing biosynthetic gene clusters.

Anti-inflammatory Quinoline Alkaloids from the Roots of Waltheria indica.

Sixteen new quinoline alkaloids (1a-7, 8a, 9, 10, 13-15, 17, and 21) and 10 known analogs (8b, 11, 12, 16, 18-20, and 22-24), along with three known cyclopeptide alkaloids (25-27), were isolated from the roots of Waltheria indica. The structures of the new compounds were elucidated by detailed NMR and circular dichroism with computational support and mass spectrometry data interpretation. Anti-inflammatory potential of isolates was evaluated based on inhibition of lipopolysaccharide (LPS)-induced nitric oxide (NO) production and tumor necrosis factor-alpha (TNF-α)-induced nuclear factor kappa B (NF-κB) activity with cell culture models. In the absence of cell growth inhibition, compounds 6, 8a, 9-11, 13, 21, and 24 reduced TNF-α-induced NF-κB activity with IC50 values ranging from 7.1 to 12.1 µM, comparable to the positive control (BAY 11-7082, IC50 = 9.7 µM). Compounds 6, 8a, 8b, and 11 showed significant NO-inhibitory activity with IC50 values ranging from 11.0 to 12.8 µM, being more active than the positive control (l-NMMA, IC50 = 22.7 µM). Structure-activity relationships indicated that NO inhibitory activity was significantly affected by C-8 substitution. Inhibition of LPS-induced nitric oxide synthase (iNOS) by 8b [(5S)-waltherione M, IC50 11.7 ± 0.8 µM] correlated with inhibition of iNOS mRNA expression. The biological potential of W. indica metabolites supports the traditional use of this plant for the treatment of inflammatory-related disorders.

On TCR binding predictors failing to generalize to unseen peptides.

Several recent studies investigate TCR-peptide/-pMHC binding prediction using machine learning or deep learning approaches. Many of these methods achieve impressive results on test sets, which include peptide sequences that are also included in the training set. In this work, we investigate how state-of-the-art deep learning models for TCR-peptide/-pMHC binding prediction generalize to unseen peptides. We create a dataset including positive samples from IEDB, VDJdb, McPAS-TCR, and the MIRA set, as well as negative samples from both randomization and 10X Genomics assays. We name this collection of samples TChard. We propose the hard split, a simple heuristic for training/test split, which ensures that test samples exclusively present peptides that do not belong to the training set. We investigate the effect of different training/test splitting techniques on the models' test performance, as well as the effect of training and testing the models using mismatched negative samples generated randomly, in addition to the negative samples derived from assays. Our results show that modern deep learning methods fail to generalize to unseen peptides. We provide an explanation why this happens and verify our hypothesis on the TChard dataset. We then conclude that robust prediction of TCR recognition is still far for being solved.

Synthesizing theories of human language with Bayesian program induction.

Automated, data-driven construction and evaluation of scientific models and theories is a long-standing challenge in artificial intelligence. We present a framework for algorithmically synthesizing models of a basic part of human language: morpho-phonology, the system that builds word forms from sounds. We integrate Bayesian inference with program synthesis and representations inspired by linguistic theory and cognitive models of learning and discovery. Across 70 datasets from 58 diverse languages, our system synthesizes human-interpretable models for core aspects of each language's morpho-phonology, sometimes approaching models posited by human linguists. Joint inference across all 70 data sets automatically synthesizes a meta-model encoding interpretable cross-language typological tendencies. Finally, the same algorithm captures few-shot learning dynamics, acquiring new morphophonological rules from just one or a few examples. These results suggest routes to more powerful machine-enabled discovery of interpretable models in linguistics and other scientific domains.

Coumarinolignans with Reactive Oxygen Species (ROS) and NF-κB Inhibitory Activities from the Roots of Waltheria indica.

Seven new coumarinolignans, walthindicins A-F (1a, 1b, 2-5, 7), along with five known analogs (6, 8-11), were isolated from the roots of Waltheria indica. The structures of the new compounds are determined by detailed nuclear magnetic resonance (NMR), circular dichroism (CD) with extensive computational support, and mass spectroscopic data interpretation. Compounds were tested for their antioxidant activity in Human Cervical Cancer cells (HeLa cells). Compounds 1a and 6 showed higher reactive oxygen species (ROS) inhibitory activity at 20 µg/mL when compared with other natural compound-based antioxidants such as ascorbic acid. Considering the role of ROS in nuclear-factor kappa B (NF-κB) activation, compounds 1a and 6 were evaluated for NF-κB inhibitory activity and showed a concentration-dependent inhibition in Human Embryonic Kidney 293 cells (Luc-HEK-293).

Spirovetivane- and Eudesmane-Type Sesquiterpenoids Isolated from the Culture Media of Two Cyanobacterial Strains.

As part of a study examining polar metabolites produced by cyanobacterial strains, we examined media extracts of a Calothrix sp. (strain R-3-1) and a Scytonema sp. (strain U-3-3). The cell mass of each was separated from the media, and HP20 resin was added for adsorption of secreted metabolites, a relatively unexplored area of cyanobacterial chemistry. HPLC-UV-LCMS-guided isolation led to the discovery of seven sesquiterpenoid compounds with five new, one known, and one previously isolated as the methyl ester. Through a complement of 1D and 2D NMR spectroscopic techniques, the planar structures and relative configurations of the seven compounds were elucidated. Spironostoic acid (1), 11,12-didehydrospironostoic acid (2), and 12-hydroxy-2-oxo-11-epi-hinesol (4) are spirovetivane-type compounds from R-3-1, while stigolone (5), 11R,12-dihydroxystigolone (6), and 11S,12-dihydroxystigolone (7) are three eudesmane-type compounds from U-3-3. Circular dichroism was utilized to decipher the absolute configurations of new compounds 1, 2, 4, 5, 6, and 7. Due to the structural variety observed among the spirovetivane- and eudesmane-type compounds in the literature and often a lack of clarity in how determinations were made, computational spectra and model compounds were used to support the interpretation of ECD and NMR spectra. A straightforward process to determine the configuration of these systems is presented.

Zinc limitation triggers anticipatory adaptations in Mycobacterium tuberculosis.

Mycobacterium tuberculosis (Mtb) has complex and dynamic interactions with the human host, and subpopulations of Mtb that emerge during infection can influence disease outcomes. This study implicates zinc ion (Zn2+) availability as a likely driver of bacterial phenotypic heterogeneity in vivo. Zn2+ sequestration is part of "nutritional immunity", where the immune system limits micronutrients to control pathogen growth, but this defense mechanism seems to be ineffective in controlling Mtb infection. Nonetheless, Zn2+-limitation is an environmental cue sensed by Mtb, as calprotectin triggers the zinc uptake regulator (Zur) regulon response in vitro and co-localizes with Zn2+-limited Mtb in vivo. Prolonged Zn2+ limitation leads to numerous physiological changes in vitro, including differential expression of certain antigens, alterations in lipid metabolism and distinct cell surface morphology. Furthermore, Mtb enduring limited Zn2+ employ defensive measures to fight oxidative stress, by increasing expression of proteins involved in DNA repair and antioxidant activity, including well described virulence factors KatG and AhpC, along with altered utilization of redox cofactors. Here, we propose a model in which prolonged Zn2+ limitation defines a population of Mtb with anticipatory adaptations against impending immune attack, based on the evidence that Zn2+-limited Mtb are more resistant to oxidative stress and exhibit increased survival and induce more severe pulmonary granulomas in mice. Considering that extracellular Mtb may transit through the Zn2+-limited caseum before infecting naïve immune cells or upon host-to-host transmission, the resulting phenotypic heterogeneity driven by varied Zn2+ availability likely plays a key role during early interactions with host cells.

P-Alkynyl functionalized benzazaphospholes as transmetalating agents.

Exposure of 10π-electron benzazaphosphole 1 to HCl, followed by nucleophilic substitution with the Grignard reagent BrMgCCPh afforded alkynyl functionalized 3 featuring an exocyclic -C[triple bond, length as m-dash]C-Ph group with an elongated P-C bond (1.7932(19) Å). Stoichiometric experiments revealed that treatment of trans-Pd(PEt3)2(Ar)(i) (Ar = p-Me (C) or p-F (D)) with 3 generated trans-Pd(PEt3)2(Ar)(CCPh) (Ar = p-Me (E) or p-F (F)), 5, which is the result of ligand exchange between P-I byproduct 4 and C/D, and the reductively eliminated product (Ar-C[triple bond, length as m-dash]C-Ph). Cyclic voltammetry studies showed and independent investigations confirmed 4 is also susceptible to redox processes including bimetallic oxidative addition to Pd(0) to give Pd(i) dimer 6-Pd2-(P(t-Bu)3)2 and reduction to diphosphine 7. During catalysis, we hypothesized that this unwanted reactivity could be circumvented by employing a source of fluoride as an additive. This was demonstrated by conducting a Sonogashira-type reaction between 1-iodotoluene and 3 in the presence of 10 mol% Na2PdCl4, 20 mol% P(t-Bu)Cy2, and 5 equiv. of tetramethylammonium fluoride (TMAF), resulting in turnover and the isolation of Ph-C[triple bond, length as m-dash]C-(o-Tol) as the major product.

Production and excretion of astaxanthin by engineered Yarrowia lipolytica using plant oil as both the carbon source and the biocompatible extractant.

This study aimed to develop a bioprocess using plant oil as the carbon source for lipid-assimilating yeast to produce high-value astaxanthin. Using high-oleic safflower oil as a model, efficient cell growth and astaxanthin production by the engineered Yarrowia lipolytica strain ST7403 was demonstrated, and a considerable portion of astaxanthin was found excreted into the spent oil. Astaxanthin was the predominant carotenoid in the extracellular oil phase that allowed facile in situ recovery of astaxanthin without cell lysis. Autoclaving the safflower oil medium elevated the peroxide level but it declined quickly during fermentation (reduced by 84% by day 3) and did not inhibit cell growth or astaxanthin production. In a 1.5-L fed-batch bioreactor culture with a YnB-based medium containing 20% safflower oil, and with the feeding of casamino acids, astaxanthin production reached 54 mg/L (53% excreted) in 28 days. Further improvement in astaxanthin titer and productivity was achieved by restoring leucine biosynthesis in the host, and running fed-batch fermentation using a high carbon-to-nitrogen ratio yeast extract/peptone medium containing 70% safflower oil, with feeding of additional yeast extract/peptone, to attain 167 mg/L astaxanthin (48% excreted) in 9.5 days of culture. These findings facilitate industrial microbial biorefinery development that utilizes renewable lipids as feedstocks to not only produce high-value products but also effectively extract and recover the products, including non-native ones.Key Points• Yarrowia lipolytica can use plant oil as a C-source for astaxanthin production.• Astaxanthin is excreted and accumulated in the extracellular oil phase.• Astaxanthin is the predominant carotenoid in the extracellular oil phase.• Plant oil serves as a biocompatible solvent for in situ astaxanthin extraction. Graphical abstract.

MHCflurry 2.0: Improved Pan-Allele Prediction of MHC Class I-Presented Peptides by Incorporating Antigen Processing.

Computational prediction of the peptides presented on major histocompatibility complex (MHC) class I proteins is an important tool for studying T cell immunity. The data available to develop such predictors have expanded with the use of mass spectrometry to identify naturally presented MHC ligands. In addition to elucidating binding motifs, the identified ligands also reflect the antigen processing steps that occur prior to MHC binding. Here, we developed an integrated predictor of MHC class I presentation that combines new models for MHC class I binding and antigen processing. Considering only peptides first predicted by the binding model to bind strongly to MHC, the antigen processing model is trained to discriminate published mass spectrometry-identified MHC class I ligands from unobserved peptides. The integrated model outperformed the two individual components as well as NetMHCpan 4.0 and MixMHCpred 2.0.2 on held-out mass spectrometry experiments. Our predictors are implemented in the open source MHCflurry package, version 2.0 (github.com/openvax/mhcflurry).

6-Deoxy- and 11-Hydroxytolypodiols: Meroterpenoids from the Cyanobacterium HT-58-2.

Chemical investigation of cyanobacterial strain HT-58-2, which most closely aligns with the genus Brasilomena, has led to the isolation of two compounds related to tolypodiol. The structures and absolute configuration of 6-deoxytolypodiol (1) and 11-hydroxytolypodiol (2) were elucidated by spectroscopic and spectrometric analysis. While tolypodiol previously showed anti-inflammatory activity in a mouse ear edema assay, only 2 reduced in vitro thromboxane B2 and superoxide anion (O2-) generation from Escherichia coli lipopolysaccharide-activated rat neonatal microglia to any appreciable degree.

Five Ways in Which Computational Modeling Can Help Advance Cognitive Science: Lessons From Artificial Grammar Learning.

There is a rich tradition of building computational models in cognitive science, but modeling, theoretical, and experimental research are not as tightly integrated as they could be. In this paper, we show that computational techniques-even simple ones that are straightforward to use-can greatly facilitate designing, implementing, and analyzing experiments, and generally help lift research to a new level. We focus on the domain of artificial grammar learning, and we give five concrete examples in this domain for (a) formalizing and clarifying theories, (b) generating stimuli, (c) visualization, (d) model selection, and (e) exploring the hypothesis space.

Tolyporphins L-R: Unusual Tetrapyrroles from a Brasilonema sp. of Cyanobacterium.

Tolyporphins L-R (2-8) have been isolated from a mixed cyanobacterium-microbial culture. The structures of tolyporphins L and M have been revised to four constitutional isomers, isolated as two mixtures of dioxobacteriochlorins (2/3 and 4/5). In contrast, tolyporphin P (6) is a fully oxidized tetrapyrrole, while tolyporphins Q and R (7 and 8) are oxochlorins. X-ray structures are reported for the first time for tolyporphins A (1), R (8), and E (9), revealing unexpected stereochemical variation within the series.

The Natural Products Atlas: An Open Access Knowledge Base for Microbial Natural Products Discovery.

Despite rapid evolution in the area of microbial natural products chemistry, there is currently no open access database containing all microbially produced natural product structures. Lack of availability of these data is preventing the implementation of new technologies in natural products science. Specifically, development of new computational strategies for compound characterization and identification are being hampered by the lack of a comprehensive database of known compounds against which to compare experimental data. The creation of an open access, community-maintained database of microbial natural product structures would enable the development of new technologies in natural products discovery and improve the interoperability of existing natural products data resources. However, these data are spread unevenly throughout the historical scientific literature, including both journal articles and international patents. These documents have no standard format, are often not digitized as machine readable text, and are not publicly available. Further, none of these documents have associated structure files (e.g., MOL, InChI, or SMILES), instead containing images of structures. This makes extraction and formatting of relevant natural products data a formidable challenge. Using a combination of manual curation and automated data mining approaches we have created a database of microbial natural products (The Natural Products Atlas, www.npatlas.org) that includes 24 594 compounds and contains referenced data for structure, compound names, source organisms, isolation references, total syntheses, and instances of structural reassignment. This database is accompanied by an interactive web portal that permits searching by structure, substructure, and physical properties. The Web site also provides mechanisms for visualizing natural products chemical space and dashboards for displaying author and discovery timeline data. These interactive tools offer a powerful knowledge base for natural products discovery with a central interface for structure and property-based searching and presents new viewpoints on structural diversity in natural products. The Natural Products Atlas has been developed under FAIR principles (Findable, Accessible, Interoperable, and Reusable) and is integrated with other emerging natural product databases, including the Minimum Information About a Biosynthetic Gene Cluster (MIBiG) repository, and the Global Natural Products Social Molecular Networking (GNPS) platform. It is designed as a community-supported resource to provide a central repository for known natural product structures from microorganisms and is the first comprehensive, open access resource of this type. It is expected that the Natural Products Atlas will enable the development of new natural products discovery modalities and accelerate the process of structural characterization for complex natural products libraries.

Universality and diversity in human song.

What is universal about music, and what varies? We built a corpus of ethnographic text on musical behavior from a representative sample of the world's societies, as well as a discography of audio recordings. The ethnographic corpus reveals that music (including songs with words) appears in every society observed; that music varies along three dimensions (formality, arousal, religiosity), more within societies than across them; and that music is associated with certain behavioral contexts such as infant care, healing, dance, and love. The discography-analyzed through machine summaries, amateur and expert listener ratings, and manual transcriptions-reveals that acoustic features of songs predict their primary behavioral context; that tonality is widespread, perhaps universal; that music varies in rhythmic and melodic complexity; and that elements of melodies and rhythms found worldwide follow power laws.

A thousand studies for the price of one: Accelerating psychological science with Pushkin.

Half of the world's population has internet access. In principle, researchers are no longer limited to subjects they can recruit into the laboratory. Any study that can be run on a computer or mobile device can be run with nearly any demographic anywhere in the world, and in large numbers. This has allowed scientists to effectively run hundreds of experiments at once. Despite their transformative power, such studies remain rare for practical reasons: the need for sophisticated software, the difficulty of recruiting so many subjects, and a lack of research paradigms that make effective use of their large amounts of data, due to such realities as that they require sophisticated software in order to run effectively. We present Pushkin: an open-source platform for designing and conducting massive experiments over the internet. Pushkin allows for a wide range of behavioral paradigms, through integration with the intuitive and flexible jsPsych experiment engine. It also addresses the basic technical challenges associated with massive, worldwide studies, including auto-scaling, extensibility, machine-assisted experimental design, multisession studies, and data security.

E-Selective Synthesis and Coordination Chemistry of Pyridine-Phosphaalkenes: Five Ligands Produce Four Distinct Types of Ru(II) Complexes.

Pyridine-phosphaalkene (PN) ligands 2a-e were prepared in an E-selective fashion using phospha-Wittig methodology. Treatment of these five ligands, varying only in their 6-substituent with RuCl2(PPh3)3, produced four distinct types of coordination complexes: pyridine-phosphaalkene-derived 3b,d, cyclized 4e, and six-coordinate 5a and 6c. Prolonged heating of 3b,d in THF resulted in C-H activation of the Mes* group and cyclization to give 4b,d featuring a bidentate pyridine-phospholane ligand bound to the metal center. Complex 5a, also possessing a newly formed phospholane ring, contained a different spatial arrangement of donors to Ru(II) with an agostic Ru-H-C interaction serving as the sixth donor to the transition metal center. Ligands 2b,d,e and Ru(II) complexes 3b, 4b,e and 5a were all characterized by X-ray crystallography. Six-coordinate 6c featured a structure similar to 4b,d,e, but with the CF3 substituent acting as a weakly bound sixth ligand to the Ru(II) center, as observed by 31P{1H} and19F NMR spectroscopy. The calculated structure of 6c established that the closest Ru- - -F contact was at 2.978 Å.