Biology subjects

Goldstein, A.

Publications and source records attributed to Goldstein, A..

36 records · Page 2Linked to original sources

bioRxiv2023

Deep speech-to-text models capture the neural basis of spontaneous speech in everyday conversations

Humans effortlessly use the continuous acoustics of speech to communicate rich linguistic meaning during everyday conversations. In this study, we leverage 100 hours (half a million words) of spontaneous open-ended conversations and concurrent high-quality neural activity recorded using electrocorticography (ECoG) to decipher the neural basis of real-world speech production and comprehension. Employing a deep multimodal speech-to-text model named Whisper, we develop encoding models capable of accurately predicting neural responses to both acoustic and semantic aspects of speech. Our encoding models achieved high accuracy in predicting neural responses in hundreds of thousands of words across many hours of left-out recordings. We uncover a distributed cortical hierarchy for speech and language processing, with sensory and motor regions encoding acoustic features of speech and higher-level language areas encoding syntactic and semantic information. Many electrodes--including those in both perceptual and motor areas--display mixed selectivity for both speech and linguistic features. Notably, our encoding model reveals a temporal progression from language-to-speech encoding before word onset during speech production and from speech-to-language encoding following word articulation during speech comprehension. This study offers a comprehensive account of the unfolding neural responses during fully natural, unbounded daily conversations. By leveraging a multimodal deep speech recognition model, we highlight the power of deep learning for unraveling the neural mechanisms of language processing in real-world contexts.

neuroscience↗

bioRxiv2023

Cardiomyocyte apoptosis contributes to contractile dysfunction in stem cell model of MYH7 E848G hypertrophic cardiomyopa-thy

Missense mutations in myosin heavy chain 7 (MYH7) are a common cause of hyper-trophic cardiomyopathy (HCM), but the molecular mechanisms underlying MYH7-based HCM remain unclear. In this work, we generated cardiomyocytes derived from isogenic human induced pluripotent stem cells to model the heterozygous pathogenic MYH7 missense variant, E848G, which is associated with left ventricular hypertrophy and adultonset systolic dysfunction. MYH7E848G/+ increased cardiomyocyte size and reduced the maximum twitch forces of engineered heart tissue, consistent with the systolic dysfunction in MYH7 E848G HCM patients. Interestingly, MYH7E848G/+ cardiomyocytes more frequently underwent apoptosis that was associated with increased p53 activity relative to controls. However, genetic ablation of TP53 did not rescue cardiomyocyte survival or restore engineered heart tissue twitch force, indicating MYH7E848G/+ cardiomyocyte apoptosis and contractile dysfunction are p53-independent. Overall, our findings suggest that cardiomyocyte apoptosis plays an important role in the MYH7E848G/+ HCM phenotype in vitro and that future efforts to target p53-independent cell death pathways may be beneficial for the treatment of HCM patients with systolic dysfunction.

molecular biology↗

bioRxiv2022

Correspondence between the layered structure of deep language models and temporal structure of natural language processing in the human brain

Deep language models (DLMs) provide a novel computational paradigm for how the brain processes natural language. Unlike symbolic, rule-based models described in psycholinguistics, DLMs encode words and their context as continuous numerical vectors. These "embeddings" are constructed by a sequence of computations organized in "layers" to ultimately capture surprisingly sophisticated representations of linguistic structures. How does this layered hierarchy map onto the human brain during natural language comprehension? In this study, we used electrocorticography (ECoG) to record neural activity in language areas along the superior temporal gyrus and inferior frontal gyrus while human participants listened to a 30-minute spoken narrative. We supplied this same narrative to a high-performing DLM (GPT2-XL) and extracted the contextual embeddings for each word in the story across all 48 layers of the model. We next trained a set of linear encoding models to predict the temporally-evolving neural activity from the embeddings at each layer. We found a striking correspondence between the layer-by-layer sequence of embeddings from GPT2-XL and the temporal sequence of neural activity in language areas. In addition, we found evidence for the gradual accumulation of recurrent information along the linguistic processing hierarchy. However, we also noticed additional neural processes in the brain, but not in DLMs, during the processing of surprising (unpredictable) words. These findings point to a connection between human language processing and DLMs where the layer-by-layer accumulation of contextual information in DLM embeddings matches the temporal dynamics of neural activity in high-order language areas.

neuroscience↗

bioRxiv2022

Microwell-based flow culture reduces hypoxia-induced phenotype and restores drug response in prostate cancer spheroids

3D cancer spheroids represent a highly promising model for study of cancer progression and therapeutic development. Wide-scale adoption of cancer spheroids, however, remains a challenge due to the lack of control over hypoxic gradients that may cloud the assessment of cell morphology and drug response. Here, we present a Microwell Flow Device (MFD) that generates in-well laminar flow around 3D tissues via repetitive tissue sedimentation. Using a prostate cancer cell line, we demonstrate the spheroids in the MFD exhibit improved cell growth, reduced necrotic core formation, enhanced structural integrity, and down-regulated expression of cell stress genes. The flow-cultured spheroids also exhibit an improved sensitivity to chemotherapy with greater transcriptional response. These results demonstrate how fluidic stimuli reveal the cellular phenotype previously masked by severe necrosis. Our platform advances 3D cellular models and enables study into hypoxia modulation, cancer metabolism, and drug screening within pathophysiological conditions.

bioengineering↗

bioRxiv2022

Reconstructing the cascade of language processing in the brain using the internal computations of a transformer-based language model

Humans use complex linguistic structures to transmit ideas to one another. The brain is thought to deploy specialized computations to process these structures. Recently, a new class of artificial neural networks based on the Transformer architecture has revolutionized the field of language modeling, attracting attention from neuroscientists seeking to understand the neurobiology of language in silico. Transformers integrate information across words via multiple layers of structured circuit computations, forming increasingly contextualized representations of linguistic content. Prior work has focused on the internal representations (the "embeddings") generated by these circuits. In this paper, we instead analyze the circuit computations directly: we deconstruct these computations into functionally-specialized "transformations" to provide a complementary window onto linguistic computations in the human brain. Using functional MRI data acquired while participants listened to naturalistic spoken stories, we first verify that the transformations account for considerable variance in brain activity across the cortical language network. We then demonstrate that the emergent syntactic computations performed by individual, functionally-specialized "attention heads" differentially predict brain activity in specific cortical regions. These heads fall along gradients corresponding to different layers, contextual distances, and syntactic dependencies in a low-dimensional cortical space. Our findings indicate that large language models and the cortical language network may converge on similar trends of functional specialization for processing natural language.

neuroscience↗

bioRxiv2022

Brain embeddings with shared geometry to artificial contextual embeddings, as a code for representing language in the human brain

Contextual embeddings, derived from deep language models (DLMs), provide a continuous vectorial representation of language. This embedding space differs fundamentally from the symbolic representations posited by traditional psycholinguistics. Do language areas in the human brain, similar to DLMs, rely on a continuous embedding space to represent language? To test this hypothesis, we densely recorded the neural activity in the Inferior Frontal Gyrus (IFG, also known as Brocas area) of three participants using dense intracranial arrays while they listened to a 30-minute podcast. From these fine-grained spatiotemporal neural recordings, we derived for each patient a continuous vectorial representation for each word (i.e., a brain embedding). Using stringent, zero-shot mapping, we demonstrated that brain embeddings in the IFG and the DLM contextual embedding space have strikingly similar geometry. This shared geometry allows us to precisely triangulate the position of unseen words in both the brain embedding space (zero-shot encoding) and the DLM contextual embedding space (zero-shot decoding). The continuous brain embedding space provides an alternative computational framework for how natural language is represented in cortical language areas.

neuroscience↗

bioRxiv2022

How to resolve conflicting conservation objectives: A decision support tool for the global selection of multi-purpose protected areas

The establishment and maintenance of protected areas (PAs) is viewed as a key action in delivering post-2020 biodiversity targets. PAs often need to meet multiple objectives, ranging from biodiversity protection to ecosystem service provision and climate change mitigation, but available land and conservation funding is limited. Therefore, optimizing resources by selecting the most beneficial PAs is vital. Here, we advocate for a flexible and transparent approach to selecting protected areas based on multiple objectives, and illustrate this with a decision support tool on a global scale. The tool allows weighting and prioritization of different conservation objectives according to user-specified preferences, as well as real-time comparison of the selected areas that result from such different priorities. We apply the tool across 1347 terrestrial PAs and highlight frequent trade-offs among different objectives, e.g., between species protection and ecosystem integrity. Outputs indicate that decision makers frequently face trade-offs among conflicting objectives. Nevertheless, we show that transparent decision-support tools can reveal synergies and trade-offs associated with PA selection, thereby helping to illuminate and resolve land-use conflicts embedded in divergent societal and political demands and values.

ecology↗

bioRxiv2021

Fine-tuning of deep language models as a computational framework of modeling listeners' perspective during language comprehension

Computational Deep Language Models (DLMs) have been shown to be effective in predicting neural responses during natural language processing. This study introduces a novel computational framework, based on the concept of fine-tuning (Hinton, 2007), for modeling differences in interpretation of narratives based on the listeners perspective (i.e. their prior knowledge, thoughts, and beliefs). We draw on an fMRI experiment conducted by Yeshurun et al. (2017), in which two groups of listeners were listening to the same narrative but with two different perspectives (cheating versus paranoia). We collected a dedicated dataset of ~3000 stories, and used it to create two modified (fine-tuned) versions of a pre-trained DLM, each representing the perspective of a different group of listeners. Information extracted from each of the two fine-tuned models was better fitted with neural responses of the corresponding group of listeners. Furthermore, we show that the degree of difference between the listeners interpretation of the story - as measured both neurally and behaviorally - can be approximated using the distances between the representations of the story extracted from these two fine-tuned models. These models-brain associations were expressed in many language-related brain areas, as well as in several higher-order areas related to the default-mode and the mentalizing networks, therefore implying that computational fine-tuning reliably captures relevant aspects of human language comprehension across different levels of cognitive processing.

neuroscience↗

bioRxiv2021

An international report on bacterial communities in esophageal squamous cell carcinoma

The incidence of esophageal squamous cell carcinoma (ESCC) is disproportionately high in the eastern corridor of Africa and parts of Asia. Emerging research has identified a potential association between poor oral health and ESCC. One proposed biological pathway linking poor oral health and ESCC involves the alteration of the microbiome. Thus, we performed an integrated analysis of four independent sequencing efforts of ESCC tumors from patients from high- and low-incidence regions of the world. Using whole genome sequencing (WGS) and RNA sequencing (RNAseq) of ESCC tumors and WGS of synchronous collections of saliva specimens from 61 patients in Tanzania, we identified a community of bacteria, including members of the genera Fusobacterium, Selenomonas, Prevotella, Streptococcus, Porphyromonas, Veillonella, and Campylobacter, present at high abundance in ESCC tumors. We then characterized the microbiome of 238 ESCC tumor specimens collected in two additional independent sequencing efforts consisting of patients from other high-ESCC incidence regions (Tanzania, Malawi, Kenya, Iran, China). This analysis revealed a similar tumor enrichment of the ESCC-associated bacterial community in these cancers. Because these genera are traditionally considered members of the oral microbiota, we explored if there is a relationship between the synchronous saliva and tumor microbiomes of ESCC patients in Tanzania. Comparative analyses revealed that paired saliva and tumor microbiomes are significantly similar with a specific enrichment of Fusobacterium and Prevotella in the tumor microbiome. Together, these data indicate that cancer-associated oral bacteria are associated with ESCC tumors at the time of diagnosis and support a model in which oral bacteria are present in high abundance in both saliva and tumors of ESCC patients. Longitudinal studies of the pre-diagnostic oral microbiome are needed to investigate whether these cross-sectional similarities reflect temporal associations.

microbiology↗

bioRxiv2020

Narratives: fMRI data for evaluating models of naturalistic language comprehension

The "Narratives" collection aggregates a variety of functional MRI datasets collected while human subjects listened to naturalistic spoken stories. The current release includes 345 subjects, 891 functional scans, and 27 diverse stories of varying duration totaling ~4.6 hours of unique stimuli (~43,000 words). This data collection is well-suited for naturalistic neuroimaging analysis, and is intended to serve as a benchmark for models of language and narrative comprehension. We provide standardized MRI data accompanied by rich metadata, preprocessed versions of the data ready for immediate use, and the spoken story stimuli with time-stamped phoneme- and word-level transcripts. All code and data are publicly available with full provenance in keeping with current best practices in transparent and reproducible neuroimaging.

neuroscience↗

bioRxiv2020

Thinking ahead: prediction in context as a keystone of language in humans and machines

Departing from traditional linguistic models, advances in deep learning have resulted in a new type of predictive (autoregressive) deep language models (DLMs). Using a self-supervised next-word prediction task, these models are trained to generate appropriate linguistic responses in a given context. We provide empirical evidence that the human brain and autoregressive DLMs share three fundamental computational principles as they process natural language: 1) both are engaged in continuous next-word prediction before word-onset; 2) both match their pre-onset predictions to the incoming word to calculate post-onset surprise (i.e., prediction error signals); 3) both represent words as a function of the previous context. In support of these three principles, our findings indicate that: a) the neural activity before word-onset contains context-dependent predictive information about forthcoming words, even hundreds of milliseconds before the words are perceived; b) the neural activity after word-onset reflects the surprise level and prediction error; and c) autoregressive DLM contextual embeddings capture the neural representation of context-specific word meaning better than arbitrary or static semantic embeddings. Together, our findings suggest that autoregressive DLMs provide a novel and biologically feasible computational framework for studying the neural basis of language.

neuroscience↗

bioRxiv2020

Phenotypic plasticity of ER+ breast cancer in the bone microenvironment

ER+ breast cancer exhibits a strong bone-tropism in metastasis. How the bone microenvironment impacts the ER signaling and endocrine therapies remains poorly understood. Here, we discover that the osteogenic niche transiently reduces ER expression and activities specifically in bone micrometastases (BMMs), leading to endocrine resistance. This is mediated by gap junctions and paracrine FGF/PDGF signaling, which together generate a stable "memory": cancer cells extracted from bone remain resistant to endocrine therapies for several generations. Using single cell-derived populations (SCPs), we demonstrated that this process is independent of clonal selection, and represents an EZH2-mediated epigenomic reprogramming. EZH2 drives ER+ BMMs toward a basal and stem-like state. EZH2 inhibition reverses endocrine resistance. Our data demonstrates how epigenomic adaptation to the bone microenvironment drives phenotypic plasticity of metastatic seeds and alters their therapeutic responses together with clonal selection, and provides insights into the clinical enigma of ER+ metastatic recurrences despite endocrine therapies.

cancer biology↗

bioRxiv2020

Stanniocalcin 1a is a Ca2+-regulated switch controlling epithelial cell quiescence-proliferation balance and Ca2+ uptake

The mechanisms governing cell quiescence-proliferation balance are poorly defined. Using a zebrafish model, here we report that Stc1a, a glycoprotein known as a hypocalcemic hormone, not only inhibits epithelial calcium uptake but also functions as a Ca2+-regulated switch controlling epithelial cell quiescence-proliferation balance. Among the 4 stc genes, only the stc1a expression is [Ca2+]-dependent. Genetic deletion of stc1a, but not stc2b, resulted in elevated body Ca2+ contents, ectopic Ca2+ deposit, body swelling, and premature death. Reducing epithelial calcium channel Trpv6-mediated Ca2+ uptake alleviated these phenotypes. Loss of Stc1a also promoted quiescent epithelial cells to re-enter the cell cycle. This action was accompanied by local IGF signaling activation and increased expression in papp-aa, a zinc metalloproteinase degrading Igfbp5a. Genetic deletion of papp-aa or igfbp5a abolished the elevated epithelial cell reactivation in stc1a-/- mutants. Likewise, inhibition of IGF1 receptor, PI3 kinase, Akt, and Tor signaling abolished epithelial cell reactivation. These results reveal that Stc1a plays dual roles in regulating epithelial calcium uptake and cell quiescence-proliferation balance and implicate Trpv6 and Papp-aa-Igfbp5a-IGF signaling in these functions.

developmental biology↗

bioRxiv2020

Highly multiplexed immune profiling throughout adulthood reveals kinetics of lymphocyte infiltration in the aging mouse prostate

Aging is a significant risk factor for cancer in several tissues, including the prostate. Defining the kinetics of age-related changes in these tissues is critical for identifying regulators of aging and evaluating interventions to slow the aging process and reduce disease risk. An altered microenvironment is characteristic of prostatic aging in mice. Whether features of aging in the prostate emerge predominantly in old age or earlier in adulthood has not previously been established. Using comprehensive immune profiling and time-course analysis, we show that populations of T and B lymphocytes increase in the mouse prostate between 6 and 12 months of age. When comparing the prostate to other urogenital tissues, we found similar features of age-related inflammation in the mouse bladder. In summary, our study offers new insight into the kinetics of prostatic inflammaging and the window when interventions to slow down age-related changes may be most effective.

immunology↗

bioRxiv2019

Robust-fit to nature: an evolutionary perspective on biological (and artificial) neural networks

Evolution is a blind fitting process by which organisms, over generations, adapt to the niches of an ever-changing environment. Does the mammalian brain use similar brute-force fitting processes to learn how to perceive and act upon the world? Recent advances in training deep neural networks has exposed the power of optimizing millions of synaptic weights to map millions of observations along ecologically relevant objective functions. This class of models has dramatically outstripped simpler, more intuitive models, operating robustly in real-life contexts spanning perception, language, and action coordination. These models do not learn an explicit, human-interpretable representation of the underlying structure of the data; rather, they use local computations to interpolate over task-relevant manifolds in a high-dimensional parameter space. Furthermore, counterintuitively, over-parameterized models, similarly to evolutionary processes, can be simple and parsimonious as they provide a versatile, robust solution for learning a diverse set of functions. In contrast to traditional scientific models, where the ultimate goal is interpretability, over-parameterized models eschew interpretability in favor of solving real-life problems or tasks. We contend that over-parameterized blind fitting presents a radical challenge to many of the underlying assumptions and practices in computational neuroscience and cognitive psychology. At the same time, this shift in perspective informs longstanding debates and establishes unexpected links with evolution, ecological psychology, and artificial life.

neuroscience↗

bioRxiv2019

Neural oscillations while remembering traumatic memories in PTSD

BackgroundIn posttraumatic stress disorder (PTSD), the traumatic event is often re-experienced through vivid sensory fragments of the traumatic experience. Though the sensory phenomenology of traumatic memories is well established, neural indications for this qualitative experience are lacking. The current study aimed at monitoring the oscillatory brain activity of PTSD patients during directed and imaginal exposure to the traumatic memory using magnetoencephalography (MEG), in a paradigm resembling exposure therapy.\n\nMethodsBrain activity of healthy trauma-exposed controls and PTSD participants was measured with MEG as they listened to individualized trauma narratives as well as to a neutral narrative and as they imagined the narrative in detail. Source localization analysis on varied frequency bands was conducted in order to map neural generators of altered oscillatory activity.\n\nResultsPTSD patients exhibited increased power of high-frequency bands over visual areas and increased delta and theta power over auditory areas in response to trauma recollection compared to neutral recollection, while controls did not show such differential activation. PTSD participants also showed abnormal modulation of lower frequencies in the medial prefrontal cortex.\n\nConclusionsElicitation of traumatic memories results in a distinct neural pattern in PTSD patients compared to healthy trauma-exposed individuals. Investigating the oscillatory neural dynamics of PTSD patients can help us better understand the processes underlying trauma re-experiencing.

neuroscience↗

bioRxiv2018

Mutations in ELAC2 associated with hypertrophic cardiomyopathy impair mitochondrial tRNA 3’-end processing

Dysfunction of mitochondrial gene expression, caused by mutations in either the mitochondrial or nuclear genomes, is associated with a diverse group of human disorders characterized by impaired mitochondrial respiration. Within this group, an increasing number of mutations have been identified in nuclear genes involved in mitochondrial RNA metabolism. For instance, pathogenic mutations have been identified in the genes encoding enzymes involved in the precursor transcript processing, including ELAC2. The ELAC2 gene codes for the mitochondrial RNase Z, which is responsible for endonucleolytic cleavage of the 3 ends of mitochondrial pre-tRNAs. Here, we report the identification of sixteen novel ELAC2 variants in individuals presenting with mitochondrial respiratory chain deficiency, hypertrophic cardiomyopathy and lactic acidosis. We provided further evidence for the pathogenicity of the three previously reported variants by studying the RNase Z activity in an in vitro system and applied this recombinant system to investigate all novel missense variants, confirming the pathogenic role of these new ELAC2 mutations. We also modelled the residues affected by missense mutation in solved RNase Z structures, providing insight into enzyme structure and function. Finally, we show that primary fibroblasts from the individuals with novel ELAC2 variants have elevated levels of unprocessed mitochondrial RNA precursors. Our study thus broadly confirms the correlation of ELAC2 variants with severe infantile-onset forms of hypertrophic cardiomyopathy and mitochondrial respiratory chain dysfunction. One rare missense variant associated with the occurrence of prostate cancer (p.Arg781His) impairs the mitochondrial RNase Z activity of ELAC2, possibly indicating a functional link between tumorigenesis and mitochondrial RNA metabolism.

genetics↗

bioRxiv2018

Synthetic-evolution reveals that phosphoregulation of the mitotic kinesin-5 Cin8 is constrained

Cdk1 has been found to phosphorylate the majority of its substrates in disordered regions. These phosphorylation sites do not appear to require precise positioning for their function. The mitotic kinesin-5 Cin8 was shown to be phosphoregulated at three Cdk1 sites in disordered loops within its catalytic motor domain. Here, we examined the flexibility of Cin8 phosphoregulation by analyzing the phenotypes of synthetic Cdk1-sites that were systematically generated by single amino-acid substitutions, starting from a phosphodeficient variant of Cin8. Out of 29 synthetic Cdk1 sites that we created, eight were non-functional; 19 were neutral, similar to the phosphodeficient variant; and two gave rise to phosphorylation-dependent spindle phenotypes. Of these two, one site resulted in novel phosphoregulation, and only one site, immediately adjacent to a native Cdk1 site, produced phosphoregulation similar to wild-type. This study shows that, while the gain of a single phosphorylation site can confer regulation and modulate the dynamics of the spindle, to achieve optimal regulation of a mitotic kinesin-5 motor protein, phosphoregulation has to be site-specific and precise.

cell biology↗