Biology subjects

Thomas, N.

Publications and source records attributed to Thomas, N..

32 records · Page 2Linked to original sources

bioRxiv2021

End-to-end learning of multiple sequence alignments with differentiable Smith-Waterman

Multiple Sequence Alignments (MSAs) of homologous sequences contain information on structural and functional constraints and their evolutionary histories. Despite their importance for many downstream tasks, such as structure prediction, MSA generation is often treated as a separate pre-processing step, without any guidance from the application it will be used for. Here, we implement a smooth and differentiable version of the Smith-Waterman pairwise alignment algorithm that enables jointly learning an MSA and a downstream machine learning system in an end-to-end fashion. To demonstrate its utility, we introduce SMURF (Smooth Markov Unaligned Random Field), a new method that jointly learns an alignment and the parameters of a Markov Random Field for unsupervised contact prediction. We find that SMURF learns MSAs that mildly improve contact prediction on a diverse set of protein and RNA families. As a proof of concept, we demonstrate that by connecting our differentiable alignment module to AlphaFold and maximizing predicted confidence, we can learn MSAs that improve structure predictions over the initial MSAs. Interestingly, the alignments that improve AlphaFold predictions are self-inconsistent and can be viewed as adversarial. This work highlights the potential of differentiable dynamic programming to improve neural network pipelines that rely on an alignment and the potential dangers of relying on black-box methods for optimizing predictions of protein sequences.

bioinformatics↗

bioRxiv2021

Shared and distinct genetic features in human and canine B-cell lymphomas

Animal models of human cancers are an important tool for the development and preclinical evaluation of therapeutics. Canine B-cell lymphoma (cBCL) is an appealing model for human mature B-cell neoplasms due to the high sequence similarity in cancer genes to humans and inactive telomerase in adult tissues. We performed targeted sequencing on 86 canine patients from the Canine Comparative Oncology Genomic Consortium, with 61 confirmed as B-cell lymphomas. We confirmed a high frequency of mutations in TRAF3 (45%) and FBXW7 (20%) as has been reported by our group and others. We also note a higher frequency of DDX3X (20%) and MYC (13%) mutations in our canine cohort. We compared the pattern and incidence of mutations in cBCL to human diffuse large B-cell lymphoma (hDLBCL) and human Burkitt lymphoma (hBL). Canine MYC mutations displayed a focal pattern with 80% of mutations affecting the conserved phosphodegron sequence in MYC box 1, which are known to stabilize MYC protein. We also note that MYC and FBXW7 mutations do not co-occur in our cBCL cohort, leading to the hypothesis that these mutations represent alternative approaches to stabilize MYC in canine lymphoma. We observed striking differences in the pattern of DDX3X mutations in canine lymphoma as compared to hBL and uncovered a sex-specific pattern of DDX3X mutations in hBL that is not consistent with those identified in canine lymphomas. In sum, we describe key differences between cBCL and human mature B-cell lymphomas which may indicate differences in the biology of these cancers. This should be considered in future studies of cBCL as a model of human lymphomas.

cancer biology↗

bioRxiv2021

Non-coding NFKBIZ 3' UTR mutations promote cell growth and resistance to targeted therapies in diffuse large B-cell lymphoma

Amplifications and non-coding 3' UTR mutations affecting NFKBIZ have been identified as recurrent genetic events in diffuse large B-cell lymphoma (DLBCL). We confirm the prevalence and pattern of NFKBIZ 3' UTR mutations in independent cohorts and determine they are enriched in the ABC subtype as well as the recently described novel BN2/C1/NOTCH2 classes of DLBCL. Presently, the effects of and mechanism by which non-coding mutations can act as cancer drivers has been relatively unexplored. Here, we provide a functional characterization of these non-coding NFKBIZ 3' UTR mutations. We demonstrate that the resulting elevated expression of I{kappa}B-{zeta} confers growth advantage in DLBCL cell lines and primary germinal center B-cells as well as nominate novel I{kappa}B-{zeta} target genes with potential therapeutic implications. The limited responses to targeted treatments in DLBCL, particularly those targeting the NF-{kappa}B axis, led us to investigate and confirm that NFKBIZ 3' UTR mutations affect response to therapeutics and suggest it may be a useful predictive biomarker. Statement of SignificanceThrough functional characterization we reveal that non-coding NFKBIZ 3' UTR mutations are a common driver in DLBCL, and mutation status may be a relevant biomarker to predict poor response to therapeutics targeting the NF-{kappa}B pathway.

cancer biology↗

bioRxiv2021

Enhanced lipogenesis through Pparγ helps cavefish adapt to food scarcity.

Nutrient availability varies seasonally and spatially in the wild. The resulting nutrient limitation or restricted access to nutrients pose a major challenge for every organism. While many animals, such as hibernating animals, evolved strategies to overcome periods of nutrient scarcity, the cellular mechanisms of these strategies are poorly understood. Cave environments represent an extreme example of nutrient deprived environments since the lack of sunlight and therefore primary energy production drastically diminishes the nutrient availability. Here, we used Astyanax mexicanus, which includes river-dwelling surface fish and cave adapted cavefish populations to study the genetic adaptation to nutrient limitations. We show that cavefish populations store large amounts of fat in different body regions when fed ad libitum in the lab. We found higher expression of lipogenesis genes in cavefish livers when fed the same amount of food as surface fish, suggesting an improved ability of cavefish to use lipogenesis to convert available energy into triglycerides for storage into adipose tissue. Moreover, the lipid metabolism regulator, Peroxisome proliferator-activated receptor {gamma} (Ppar{gamma}), is upregulated at both transcript and protein levels in cavefish livers. Chromatin Immunoprecipitation sequencing (ChIP seq) showed that Ppar{gamma} binds cavefish promoter regions of genes to a higher extent than surface fish. Finally, we identified two possible regulatory mechanisms of Ppar{gamma} in cavefish: higher amounts of ligands of the nuclear receptor, and nonsense mutations in per2, a known repressor of Ppar{gamma}. Taken together, our study reveals that upregulated Ppar{gamma} promotes higher levels of lipogenesis in the liver and contributes to higher body fat accumulation in cavefish populations, an important adaptation to nutrient limited environments.

evolutionary biology↗

bioRxiv2021

Direct Nanopore Sequencing of Individual Full Length tRNA Strands

We describe a method for direct tRNA sequencing using the Oxford Nanopore MinION. The principal technical advance is custom adapters that facilitate end-to-end sequencing of individual tRNA molecules. A second advance is a Nanopore sequencing pipeline optimized for tRNA. We tested this method using purified E. coli tRNAfMet, tRNALys, and tRNAPhe samples. 76-to-92% of individual aligned tRNA sequence reads were full length. As proof of concept, we showed that Nanopore sequencing detected all 42 expected tRNA isoacceptors in total E. coli tRNA. Alignment-based comparison between the three purified tRNAs and their synthetic controls revealed systematic miscalls at or adjacent to the positions of known nucleotide modifications. Systematic miscalls were also observed proximal to known modifications in total E. coli tRNA alignments.

bioengineering↗

bioRxiv2021

Phospholipids stabilize binding of pituitary adenylate cyclase-activating peptide to vasoactive intestinal polypeptide receptor

Vasoactive intestinal polypeptide receptor (VIP1R) is a class B G-protein coupled receptor (GPCR) that is widely distributed throughout the central nervous system, T-lymphocytes, and peripheral tissues of organs like lungs and liver. Critical functions of these receptors render them potential pharmacological targets for the treatment of a broad spectrum of inflammatory and neurodegenerative diseases. Here we use atomistic studies to show that phospholipids can act as potent regulators of peptide binding on to the receptor. We simulated the binding of neuropeptide pituitary adenylate cyclase-activating peptide (PACAP27) into the transmembrane bundle of the receptor. The simulations reveal two lipid binding sites on the peptidic ligand for the negatively charged phosphodiester of phospholipids in the extracellular leaflet which lower the peptide-receptor binding free energy by ~8kBT. We further simulated the effect of anionic lipids phosphatidylinositol-4,5-bisphosphate (PIP2). These lipids show much stronger interaction, lowering the peptide-receptor binding energy by an additional ~7kBT compared to POPC lipids. These findings suggest that lipids can play an active role in catalyzing peptide-receptor binding and activating vasoactive intestinal polypeptide receptors.

biophysics↗

bioRxiv2021

Lateral electric field inhibits gel-to-fluid transition in lipid bilayers

We report evidence of lateral electric field-induced changes in the phase transition temperatures of lipid bilayers. Our atomic scale molecular dynamics simulations show that lateral electric field increases the melting temperature of DPPC, POPC and POPE bilayers. Remarkably, this shift in melting temperature is only induced by lateral electric field, and not normal electric field. This mechanism could provide new mechanistic insights into lipid-lipid and lipid-protein interactions in the presence of endogenous and exogenous electric fields.

biophysics↗

bioRxiv2021

Quantification of in-plane flexoelectricity in lipid bilayers

Lipid bilayers behave as 2D dielectric materials that undergo polarization and deformation in the presence of an electric field. This effect has been previously modeled by continuum theories which assume a polarization field oriented normal to the membrane surface. However, the molecular architecture of the lipids reveals that the heqadgroup dipoles are primarily oriented tangential to the membrane surface. Here, we perform atomistic and coarse-grained molecular dynamics simulations to quantify the in-plane polarization undergone by a flat bilayer and a spherical vesicle in the presence of an applied electric field. We use these predictions to compute an effective in-plane flexoelectric coefficient for four different lipid types. Our findings provide the first molecular proof of the in-plane polarization undergone by lipid bilayers and furnish the material parameter required to quantify membrane-electric field interactions.

biophysics↗

bioRxiv2020

Single Layers of Attention Suffice to Predict Protein Contacts

AO_SCPLOWBSTRACTC_SCPLOWThe established approach to unsupervised protein contact prediction estimates co-evolving positions using undirected graphical models. This approach trains a Potts model on a Multiple Sequence Alignment, then predicts that the edges with highest weight correspond to contacts in the 3D structure. On the other hand, increasingly large Transformers are being pretrained on protein sequence databases but have demonstrated mixed results for downstream tasks, including contact prediction. This has sparked discussion about the role of scale and attention-based models in unsupervised protein representation learning. We argue that attention is a principled model of protein interactions, grounded in real properties of protein family data. We introduce a simplified attention layer, factored attention, and show that it achieves comparable performance to Potts models, while sharing parameters both within and across families. Further, we extract contacts from the attention maps of a pretrained Transformer and show they perform competitively with the other two approaches. This provides evidence that large-scale pretraining can learn meaningful protein features when presented with unlabeled and unaligned data. We contrast factored attention with the Transformer to indicate that the Transformer leverages hierarchical signal in protein family databases not captured by our single-layer models. This raises the exciting possibility for the development of powerful structured models of protein family databases.1

bioinformatics↗

bioRxiv2020

Supraphysiological testosterone induces ferroptosis and activates NF-kappaB mediated immune pathways in prostate cancer through nucleophagy

The discovery that androgens play an important role in the progression of prostate cancer (PCa) has led to the development of androgen deprivation therapy as a first line of treatment against PCa. However, paradoxical growth inhibition has been observed, both experimentally and clinically, in a subset of PCa upon administration of supraphysiological levels of testosterone (SupraT). Here we report that SupraT activates cytoplasmic nucleic acid sensors and induces growth inhibition of SupraT-sensitive PCa cells. This is initiated by induction of two parallel autophagy-mediated processes, namely, ferritinophagy and nucleophagy. Consequently, autophagosomal DNA activates nucleic acid sensors that converge on NF-kappaB to drive immune signaling pathways. Chemokines and cytokines secreted by the tumor cells in response to SupraT results in increased migration of cytotoxic immune cells to tumor beds of animal xenografts and patient tumors. Collectively, our findings indicate that SupraT may inhibit a subset of PCa by activating nucleic acid sensors and downstream immune signaling.

cancer biology↗

bioRxiv2020

Electromechanics of lipid-modulated gating of Kv channels

Experimental studies reveal that anionic lipid POPA and non-phospholipid cholesterol inhibit the gating of voltage-sensitive potassium (Kv) channels at 5-10% molar concentrations. Intriguingly, other anionic lipids similar to POPA, like POPG, have minimal impact on the gating of the same channels for reasons that remain obscure. Our long-timescale atomistic simulations show that POPA preferentially solvates the voltage sensor domains of Kv channels by direct electrostatic interactions between the positively charged arginine and negatively charged phosphate groups. Cholesterol solvates the voltage sensor domains through CH-{pi} interactions between the cholesterol rings and the aromatic side chains of phenylalanine and tyrosine residues. A continuum electromechanical model predicts that POPA lipids may restrict the vertical motion of voltage-sensor domain through direct electrostatic interactions, while cholesterol may oppose the radial motion of the pore domain of the channel by increasing the mechanical rigidity of the membrane. The electromechanical model predictions are consistent with measurements of the activation curves of Kv channels for various lipids. The atomistic simulations also suggest that the solvation due to POPG is much weaker likely due to its bigger head-group size. Thus the channel activity appears to be tied to the local lipid environment, allowing lipids to regulate channel gating in low concentrations.

biophysics↗

bioRxiv2019

Coding and regulatory drivers of mantle cell lymphoma identified through exome and genome sequencing

Mantle cell lymphoma (MCL) is an uncommon B-cell non-Hodgkin lymphoma (NHL) that is incurable with standard therapies. The genetic drivers of this cancer have not been firmly established and the features known to contribute to differences in clinical course remain limited. To extend our understanding of the biological pathways involved in this malignancy, we performed a large-scale genomic analysis of MCL using data from 51 exomes alongside previously published exome cohorts. To confirm our findings, we re-sequenced the genes identified in the exome cohort in 212 MCL tumors, each having clinical follow-up data. We confirmed the prognostic association of TP53 and NOTCH1 mutations and further nominate two additional genes, EWSR1 and MEF2B, whose mutation respectively associated with poor and good outcome. Our sequencing revealed novel recurrent mutations including a unique missense hot spot in MEF2B and a pattern of non-coding mutations surrounding a single exon of the HNRNPH1 gene. We sequenced the whole genomes of 34 MCLs to confirm the focal nature of HNRNPH1 mutations. Using RNA-seq data from 110 of these cases, we identified a functional role for recurrent non-coding HNRNPH1 mutations in disrupting an auto-regulatory feedback mechanism. Overall, we identified three novel MCL-related genes with roles in RNA trafficking or splicing, namely DAZAP1, EWSR1, and HNRNPH1. Taken together, these data strongly implicate a role for aberrant regulation of splicing in MCL pathobiology.\n\nKey pointsO_LIRNA-binding proteins with roles in regulating alternative splicing, DAZAP1, EWSR1, HNRNPH1, are frequently mutated in MCL\nC_LIO_LIThe majority of recurrent somatic HNRNPH1 mutations are intronic and HNRNPH1 exhibits self-regulation through alternative splicing\nC_LI

genomics↗

bioRxiv2019

Evaluating Protein Transfer Learning with TAPE

Protein modeling is an increasingly popular area of machine learning research. Semi-supervised learning has emerged as an important paradigm in protein modeling due to the high cost of acquiring supervised protein labels, but the current literature is fragmented when it comes to datasets and standardized evaluation techniques. To facilitate progress in this field, we introduce the Tasks Assessing Protein Embeddings (TAPE), a set of five biologically relevant semi-supervised learning tasks spread across different domains of protein biology. We curate tasks into specific training, validation, and test splits to ensure that each task tests biologically relevant generalization that transfers to real-life scenarios. We bench-mark a range of approaches to semi-supervised protein representation learning, which span recent work as well as canonical sequence learning techniques. We find that self-supervised pretraining is helpful for almost all models on all tasks, more than doubling performance in some cases. Despite this increase, in several cases features learned by self-supervised pretraining still lag behind features extracted by state-of-the-art non-neural techniques. This gap in performance suggests a huge opportunity for innovative architecture design and improved modeling paradigms that better capture the signal in biological sequences. TAPE will help the machine learning community focus effort on scientifically relevant problems. Toward this end, all data and code used to run these experiments are available at https://github.com/songlab-cal/tape.

synthetic biology↗

bioRxiv2019

Comparison of vibrotactile and joint-torque feedback in a myoelectric upper-limb prosthesis

BackgroundDespite the technological advancements in myoelectric prostheses, body-powered prostheses remain a popular choice for amputees, in part due to the natural sensory advantage they provide. Research on haptic feedback in myoelectric prostheses has delivered mixed results. Furthermore, there is limited research comparing various haptic feedback modalities in myoelectric prostheses. In this paper, we present a comparison of the feedback intrinsically present in body-powered prostheses (joint-torque feedback) to a commonly proposed feedback modality for myoelectric prostheses (vibrotactile feedback). In so doing, we seek to understand whether the advantages of kinesthetic feedback present in body-powered prostheses translate to myoelectric prostheses, and whether there are differences between kinesthetic and cutaneous feedback in prosthetic applications.\n\nMethodsWe developed an experimental testbed that features a cable-driven, voluntary-closing 1-DoF prosthesis, a capstan-driven elbow exoskeleton, and a vibrotactile actuation unit. The system can present grip force to users as either a flexion moment about the elbow or vibration on the wrist. To provide an equal comparison of joint-torque and vibrotactile feedback, a stimulus intensity matching scheme was utilized. Non-amputee participants (n=12) were asked to discriminate objects of varying stiffness with the prosthesis in three conditions: no haptic feedback, vibrotactile feedback, and joint-torque feedback.\n\nResultsResults indicate that haptic feedback increased discrimination accuracy over no haptic feedback, but the difference between joint-torque feedback and vibrotactile feedback was not significant. In addition, our results highlight nuanced differences in performance depending on the objects stiffness, and suggest that participants likely pay less attention to incidental cues with the addition of haptic feedback.\n\nConclusionEven when haptic feedback is not modality matched to the task, such as in the case of vibrotactile feedback, performance with a myoelectric prosthesis can improve significantly. This implies it is possible to achieve the same benefits with vibrotactile feedback, which is cheaper and easier to implement than other forms of feedback.

bioengineering↗