LitCovid-sentences  

PMC:7371427 / 56301-83134 JSONTXT 12 Projects

Annnotations TAB TSV DIC JSON TextAE

Id Subject Object Predicate Lexical cue
T308 0-21 Sentence denotes Materials and methods
T309 23-97 Sentence denotes Unstructured biomedical knowledge synthesis and triangulation capabilities
T310 98-281 Sentence denotes In order to capture biomedical literature-based associations, the nferX platform defines two scores: a ‘local score’ and a ‘global score’, as described previously (Park et al., 2020).
T311 282-618 Sentence denotes Briefly, the local score is obtained from applying a traditional natural language processing technique which captures the strength of association between two concepts in a selected corpus of biomedical literature based on the frequency of their co-occurrence normalized by the frequency of each individual concept throughout the corpus.
T312 619-809 Sentence denotes A higher local score between Concept X and Concept Y indicates that these concepts are frequently mentioned in close proximity to each other more frequently than would be expected by chance.
T313 810-957 Sentence denotes The global score, on the other hand, is based on the neural network renaissance that has recently taken place in Natural Language Processing (NLP).
T314 958-1088 Sentence denotes To compute global scores, all tokens (e.g. words and phrases) are projected in a high-dimensional vector space of word embeddings.
T315 1089-1188 Sentence denotes These vectors serve to represent the ‘neighborhood’ of concepts which occur around a given concept.
T316 1189-1414 Sentence denotes The cosine similarity between any two vectors measures the similarity of these neighborhoods and is the basis for our global score metric, where concepts which are more similar in this vector space have a higher global score.
T317 1415-1655 Sentence denotes While the global scores in this work are computed in the embedding space of word2vec model, it can also be computed in the embedding space of any deep learning model including recent transformer-based models like BERT (Devlin et al., 2019).
T318 1656-1817 Sentence denotes These may have complementary benefits to word2vec embeddings since the embeddings are context sensitive having different vectors for different sentence contexts.
T319 1818-2056 Sentence denotes However, despite the context sensitive nature of BERT embeddings a global score computation for a phrase may still be of value given the score is computed across sentence embeddings capturing the context sensitive nature of those phrases.
T320 2057-2283 Sentence denotes From a visualization perspective, the local score and global score (‘Signals’) are represented in the platform using bubbles where bubble size corresponds to the local score and color intensity corresponds to the global score.
T321 2284-2409 Sentence denotes This allows users to rapidly determine the strength of association between any two concepts throughout biomedical literature.
T322 2410-2568 Sentence denotes We consider concepts which show both high local and global scores to be ‘concordant’ and have found that these typically recapitulate well-known associations.
T323 2569-2722 Sentence denotes One key aspect of the nferX platform is that it allows the user to query associated concepts for a virtually unbounded number of possible query concepts.
T324 2723-2765 Sentence denotes This is achieved by means of two features:
T325 2766-3018 Sentence denotes Firstly, the nferX platform allows the user to compose queries using the logical AND, OR and NOT operators to logically combine any number of biomedical concepts in a query, each combination amounting to a gross or nuanced composite biomedical concept.
T326 3019-3521 Sentence denotes Secondly, since logical combinations yield a virtually unbounded number of biomedical concepts that can be queries, the nferX platform implements a completely dynamic method of computing local scores on the fly by using novel high performance parallel and distributed algorithms that, in real time, scan hundreds of millions of documents to quickly locate user query related text fragments and count co-occurring biomedical concepts for computing strength of association scores and their significances.
T327 3522-3870 Sentence denotes The platform further leverages statistical inference to calculate ‘enrichments’ based on structured data, thus enabling real-time triangulation of signals from the unstructured biomedical knowledge graph various other structured databases (e.g. curated ontologies, RNA-sequencing datasets, human genetic associations, protein-protein interactions).
T328 3871-4042 Sentence denotes This facilitates unbiased hypothesis-free learning and faster pattern recognition, and it allows users to more holistically determine the veracity of concept associations.
T329 4043-4249 Sentence denotes Finally, the platform allows the user to identify and further examine the documents and textual fragments from which the knowledge synthesis signals are derived using the Documents and Signals applications.
T330 4251-4269 Sentence denotes Association scores
T331 4270-4602 Sentence denotes Having a method that automatically consumes a corpus and computes a numeric score that captures the strength of the association between any pair of entities is obviously beneficial because then given any entity, its association strength score with all other entities can be sorted to find a sorted list of other associated entities.
T332 4603-4727 Sentence denotes The number of times two entities mutually co-occur in ‘small’ vicinities of a corpus is the basis of all association scores.
T333 4728-4911 Sentence denotes One popular traditional measure for association strength between tokens in text is pointwise mutual information, or PMI (Evert, 2005), which we consider in several association scores.
T334 4913-4936 Sentence denotes Measures of association
T335 4937-5067 Sentence denotes Formally, an association score is some real-valued function S(q, t) where q is a query token/entity and t is another token/entity.
T336 5068-5151 Sentence denotes One important notion, the ‘vicinity’ of q, we formally denote as the Context of q :
T337 5152-5231 Sentence denotes The context of q are those corpus segments deemed to be ‘near’ or ‘local’ to q.
T338 5232-5513 Sentence denotes For single token queries (where q is a single entity and not a logical combination of entities) , q’s context consists of all corpus segments that are ‘windows’ formed by taking words within a distance w (usually a tunable parameter) of words from an occurrence of q in the corpus.
T339 5514-5706 Sentence denotes The dynamic adjacency engine generalizes this notion of context in a natural way to logical queries: the context for a logical q can be generalized as a certain set of fixed-length fragments. 
T340 5708-5722 Sentence denotes Co-occurrences
T341 5723-5786 Sentence denotes This is just the number of times t appears in the context of q.
T342 5788-5803 Sentence denotes Traditional PMI
T343 5804-5831 Sentence denotes This is log(p(t | q)/p(t)).
T344 5832-6088 Sentence denotes Here p(t | q) is the number of times t occurs in the context of q (ie co-occurrences of t and q) divided by the total length of all q contexts in the corpus, whereas p(t) is the number of occurrences of t in the entire corpus, divided by the corpus length.
T345 6090-6116 Sentence denotes Word2vec cosine similarity
T346 6117-6260 Sentence denotes The popular word2vec algorithm (Raj et al., 2013) generates a vector (we use 300-dimensional vector representation) for each token in a corpus.
T347 6261-6348 Sentence denotes The purpose of these vectors is usually to be used as features in downstream NLP tasks.
T348 6349-6390 Sentence denotes But they can also be used for similarity.
T349 6391-6556 Sentence denotes The original paper validates the vectors by testing them on word similarity tasks: the association score is the cosine between the vector for q and the vector for t.
T350 6557-6599 Sentence denotes This score only applies to single-token q.
T351 6601-6630 Sentence denotes Exponential mask PMI (ExpPMI)
T352 6631-6668 Sentence denotes This is our first new proposed score.
T353 6669-6751 Sentence denotes PMI treats every position in a binary way: it’s either in the context of q or not.
T354 6752-6902 Sentence denotes With a window size of say 50, a token which appears three words from a query q and a token which appears 45 words from a query q are treated the same.
T355 6903-7075 Sentence denotes We thought it might be useful to consider a measure which distinguishes positions in the context based on the number of words away that position is from an occurrence of q.
T356 7076-7161 Sentence denotes We did this by weighting the positions in the context by some weight between 0 and 1.
T357 7162-7299 Sentence denotes Our weighting is based on an exponential decay (which has some nice properties especially when we extend to the case of logical queries).
T358 7301-7312 Sentence denotes Local score
T359 7313-7348 Sentence denotes This is another new proposed score.
T360 7349-7462 Sentence denotes We find that PMI and ExpPMI can vary a lot for small samples (i.e. small numbers of co-occurrences, occurrences).
T361 7463-7602 Sentence denotes The Local Score is log(coocc) * sigmoid(PMI - 0.5), constructed to correct for this; we found that this formula too works well empirically.
T362 7604-7648 Sentence denotes Exponential mask local score (ExpLocalScore)
T363 7649-7761 Sentence denotes We apply both modifications together: the exponential mask score is log(weighted_coocc) * sigmoid(expPMI - 0.5).
T364 7762-7839 Sentence denotes Here weighted_coocc is the sum of the weights of the positions of the corpus.
T365 7841-7892 Sentence denotes Evaluation of literature-derived association scores
T366 7893-7974 Sentence denotes We need a notion of ground truth to evaluate the quality of association measures.
T367 7975-8095 Sentence denotes We use sets of known pairs of related entities versus a ‘control’ group of random pairs of entities of the same classes.
T368 8096-8139 Sentence denotes We use a few different sets of known pairs:
T369 8140-8200 Sentence denotes Disease-Gene relationships based on OMIM (Park et al., 2020)
T370 8201-8234 Sentence denotes Drug-Gene relationships (Table 1)
T371 8235-8281 Sentence denotes Drug-Disease relationships based on FDA labels
T372 8282-8318 Sentence denotes Drugs and their on-label indications
T373 8319-8358 Sentence denotes Drugs and their on-label adverse events
T374 8359-8395 Sentence denotes Logical queries for ambiguous tokens
T375 8396-8525 Sentence denotes One demonstration of the use of the logical query system is to disambiguate a token by conjoining it with a disambiguating token.
T376 8526-8699 Sentence denotes An example is clearer: the token ‘egfr’ can refer to the gene entity epidermal growth factor receptor, but also the test measure entity estimated glomerular filtration rate.
T377 8700-8819 Sentence denotes A query ‘egfr AND kidney’ should return results related to the latter meaning, while ‘egfr AND lung_cancer’ the former.
T378 8820-8917 Sentence denotes In particular, an unambiguous referent to the right entity should be highly related to the query.
T379 8918-9083 Sentence denotes So example known pairs in this data are (‘egfr AND kidney’, ‘estimated_glomerular_filtration_rate’) and (‘egfr AND lung_cancer’, ‘epidermal_growth_factor_receptor’).
T380 9084-9188 Sentence denotes We used an internal set of ~200–300 such (‘A AND B’, ‘C’) pairs (originally built up for other reasons).
T381 9189-9194 Sentence denotes Note:
T382 9195-9521 Sentence denotes One key drawback of the word2vec vector cosine similarity (Park et al., 2020; Mikolov et al., 2013b) method is its inability to get scores for logical queries as described above, because the method (Mikolov et al., 2013b) does not address the question of how to get vectors for queries that are logical combinations of tokens.
T383 9523-9541 Sentence denotes Evaluation metrics
T384 9542-9706 Sentence denotes Given a scoring method and a particular set of positive/control pairs, we get two sets of scores: one set for the positive pairs and one set for the negative pairs.
T385 9707-9717 Sentence denotes Cohen’s d:
T386 9718-9822 Sentence denotes We compute the Cohen’s d standard statistical measure of distance between two samples (Cohen’s D, 2016).
T387 9823-10010 Sentence denotes Mann-Whitney U (normalized): - The Mann-Whitney U is a nonparametric measure of distribution distance: it counts the number of transposed pairs (Contributors to Wikimedia projects, 2004).
T388 10012-10058 Sentence denotes Metrics based on training a 1-d logistic model
T389 10059-10171 Sentence denotes In this test, we are discriminating between two classes (true association/non-association) based on one feature.
T390 10172-10283 Sentence denotes We have two metrics based on fitting a 1-feature logistic curve to the data. (Figure 1—figure supplement 1A–B).
T391 10284-10296 Sentence denotes Brier score:
T392 10297-10538 Sentence denotes The Brier score is the average squared error of the logistic curve above: that is, for each labeled point, we square the vertical distance to the logistic curve, and average over all labeled points (Contributors to Wikimedia projects, 2005).
T393 10539-10567 Sentence denotes Log loss (dansbecker, 2018):
T394 10568-10667 Sentence denotes The logistic log loss is the average -log [model probability of true label] for each labeled point.
T395 10668-10724 Sentence denotes If the model is perfect at the point, it incurs no loss.
T396 10725-10770 Sentence denotes If it predicts 0.5, it incurs -log[0.5] loss.
T397 10771-10933 Sentence denotes If it predicts ‘yes’ with certainty when the answer is ‘no’ it incurs infinite loss (a logistic function never touches 0 or one so this won’t happen in our case).
T398 10934-10953 Sentence denotes Neg log percentile:
T399 10954-11040 Sentence denotes For most of the scoring rules, we also include a -log(percentile) version of the rule.
T400 11041-11113 Sentence denotes This is constructed as follows, for query q, token t, and score S(q, t):
T401 11114-11168 Sentence denotes Compute the scores S(q, t’) for q with every token t’.
T402 11169-11215 Sentence denotes Let R be the number of these that are nonzero.
T403 11216-11270 Sentence denotes Take the rank r of S(q, t) among all nonzero S(q, t’).
T404 11271-11340 Sentence denotes The neg log percentile score nlS(q, t) associated with S is -log(r/R)
T405 11341-11355 Sentence denotes We do this to:
T406 11356-11394 Sentence denotes control for differences across queries
T407 11395-11504 Sentence denotes control for differences in the shapes of the distributions that different association scoring functions take.
T408 11505-11576 Sentence denotes This procedure maps all the S(q, t’) to an Exponential(1) distribution.
T409 11577-11718 Sentence denotes We chose Exponential(1) because it is simple, intuitively reasonable and many of the scores naturally seemed to be approximately exponential.
T410 11720-11804 Sentence denotes High-dimensional word embeddings for determining the significant global associations
T411 11805-12214 Sentence denotes Figure 1—figure supplement 1C illustrates two histograms generated from a random set of vectors (in the vector space generated by the Neural Network) where one distribution represents all vector pairs whose cosine similarity is less than 0.32 (deemed ‘not strong associations’) and the other distribution represents all vector pairs whose cosine similarity is greater than 0.32 (deemed ‘strong associations’).
T412 12215-12375 Sentence denotes This can show how common a phenomenon it is to find word vector pairs that have very good cosine similarity values but yet not co-occur even once in the corpus.
T413 12376-12588 Sentence denotes The ‘cosine similarity >= 0.32’ bar at zero value suggests that roughly 11% of vector pairs whose cosine similarity where greater than 0.32 (‘strong associations’) never occurred together even once in a document.
T414 12589-12937 Sentence denotes It is also clear from the figure that albeit more of the mass of the ‘cosine similarity >= 0.32’ distribution is skewed to the right as expected (more co-occurrences and hence unsurprisingly larger cosine similarity values), there is a long tail of the ‘cosine similarity < 0.32’ distribution (very high co-occurrences but small cosine similarity).
T415 12938-13157 Sentence denotes The long tail is a direct consequence of negative sampling—where vectors corresponding to common words that co-occur quite often with significant words in a sliding window are moved away from vectors of the other words.
T416 13159-13252 Sentence denotes What does the word2vec neural network do from the perspective of Genes-Diseases associations?
T417 13253-13571 Sentence denotes One way to view the word2vec ‘black box’ operation from a Genes/Diseases perspective (cosine of <Gene, Disease> for all Genes and Diseases) is as a Transfer Function which changed the input probability distribution (pre-training randomly assigned word vectors for Genes and Diseases) to a new probability distribution.
T418 13572-13780 Sentence denotes The ‘null hypothesis’ (which seems to be well preserved in actuality in the way word2vec assigns random values to vectors initially) is the ‘green colored’ Cosine Distribution (Figure 1—figure supplement 1D).
T419 13781-14034 Sentence denotes Once word2vec training is over, the final word vectors are placed in specific positions in the 300-dimensional space so as to present the ‘blue colored’ Empirical distribution (the actual cosine similarity between <Gene, Disease> pairs that we observe).
T420 14035-14219 Sentence denotes The ‘orange curve’ is the 2-Gamma mixture (the parametric distribution that captures the ‘empirical distribution’ with just eight parameters (two alphas, two betas, 2 ts and two phis).
T421 14220-14252 Sentence denotes Observations from this analysis:
T422 14253-14361 Sentence denotes Note the ‘symmetrical’ cosine distribution after training becomes ‘Asymmetrical’ with a longer ‘right tail’.
T423 14362-14465 Sentence denotes The asymmetry is the reason why Gamma distribution worked better than say, Gaussian, for the curve fit.
T424 14466-14945 Sentence denotes The mean of the distribution gets shifted to the right after training as one would expect — the vectors during training are ‘brought together’ by parallelogram addition predominantly— explaining the shift to the right (negative sampling will cause a movement in the opposite direction, but that will disproportionately affect the ‘ultra-high frequency’ words, which get ‘more’ positively sampled and hence the 3-gamma with a bump near 0.6 happens for ultra-high frequency words).
T425 14946-15032 Sentence denotes The most interesting associations, by definition, are in the tail of the distribution.
T426 15034-15179 Sentence denotes What does varying the number of dimensions in the word2vec space do to the underlying cosine similarity distributions in a large textual corpora?
T427 15180-15395 Sentence denotes Figure 1—figure supplement 1E illustrates a cosine similarity probability density function (PDF) graph to visually describe the implementation of the word2vec-like Vector Space Model in various N-dimensional spaces.
T428 15396-15756 Sentence denotes As described in the Materials and methods section, the system is a Semantic Bio-Knowledge Graph of nodes representing the words/phrases chosen to be represented as vectors and edge weights determined by measures of Semantic Association Strength (e.g. the cosine similarity between a pair of word embeddings represented as vectors in a large dimensional space).
T429 15757-15874 Sentence denotes The cosine similarity ranges from 0 (representing no semantic association) to 1 (representing strongest association).
T430 15875-15982 Sentence denotes This metric of association can reflect the contextual similarity of the entities in the Biomedical Corpora.
T431 15983-16092 Sentence denotes The typical dimensionality used by our neural network for generating the Global Scores is n = 300 dimensions.
T432 16093-16307 Sentence denotes This is because, as can be seen in the graph, the distribution is highly peaked with most of the mass centered around 0 -- that is, a randomly chosen pair of vectors typically are orthogonal or close to orthogonal.
T433 16308-16453 Sentence denotes Furthermore, over 300 dimensions, the distributions all have sufficiently long tails with the most interesting (salient) biomedical associations.
T434 16455-16492 Sentence denotes Single-cell RNA-seq analysis platform
T435 16493-16618 Sentence denotes The objective of the single cell platform is to enable dynamic visualization and analysis of single-cell RNA-sequencing data.
T436 16619-16948 Sentence denotes Currently, there are over 30 scRNAseq studies available for analysis in the Single Cell app, including studies from human donors/patients covering tissues such as adipose tissue, blood, bone marrow, colon, esophagus, liver, lung, kidney, ovary, nasal epithelium, pancreas, placenta, prostate, retina, small intestine, and spleen.
T437 16949-17101 Sentence denotes Because no pan-tissue reference dataset yet exists for humans, we have manually selected individual studies to maximally cover the set of human tissues.
T438 17102-17269 Sentence denotes In some cases, these studies contain cells from both healthy donors and patients affected by a specified pathology such as ulcerative colitis (colon) or asthma (lung).
T439 17270-17621 Sentence denotes There are also a number of murine scRNAseq studies covering tissues including adipose tissue, airway epithelium, blood, bone marrow, brain, breast, colon, heart, kidney, liver, lung, ovary, pancreas, placenta, prostate, skeletal muscle, skin, spleen, stomach, small intestine, testis, thymus, tongue, trachea, urinary bladder, uterus, and vasculature.
T440 17622-17721 Sentence denotes Note that two of these murine studies (Tabula Muris and Mouse Cell Atlas) include ~20 tissues each.
T441 17723-17759 Sentence denotes Single-cell data processing pipeline
T442 17760-17944 Sentence denotes For each study, a counts matrix was downloaded from a public data repository such as the Gene Expression Omnibus (GEO) or the Broad Institute Single Cell Portal (Supplementary file 1).
T443 17945-18154 Sentence denotes Note that this data has not been re-processed from the raw sequencing output, and so it is likely that alignment and quantification of gene expression was performed using different tools for different studies.
T444 18155-18248 Sentence denotes In some cases, multiple complementary datasets have been generated from a single publication.
T445 18249-18328 Sentence denotes In these cases, we have generated separate entries in the Single Cell platform.
T446 18329-18337 Sentence denotes Table 1.
T447 18339-18361 Sentence denotes Results of evaluation.
T448 18362-18415 Sentence denotes Performance of approximately 2100 disease-gene pairs.
T449 18416-18512 Sentence denotes Assoc score↓ Cohen’s d (+) Mann-W U norm. (-) Logistic log loss (-) Logistic Brier score (-)
T450 18513-18551 Sentence denotes Cosine (w2v) 1.31 0.197 0.51 0.168
T451 18552-18587 Sentence denotes Raw PMI 2.07 0.0953 0.374 0.116
T452 18588-18636 Sentence denotes Raw PMI -log(pctile) 2.15 0.0947 0.355 0.111
T453 18637-18672 Sentence denotes Exp PMI 2.17 0.0897 0.356 0.109
T454 18673-18721 Sentence denotes Exp PMI -log(pctile) 2.21 0.0903 0.341 0.105
T455 18722-18766 Sentence denotes Raw Local Score 2.35 0.0828 0.312 0.0947
T456 18767-18824 Sentence denotes Raw Local Score -log(pctile) 2.28 0.0832 0.317 0.0963
T457 18825-18873 Sentence denotes Exp Local Score 2.34 0.0812 *0.301 *0.0915
T458 18874-18934 Sentence denotes Exp Local Score -log(pctile) *2.36 *0.0811 0.308 0.093
T459 18935-18972 Sentence denotes log(coocc) 2.24 0.097 0.348 0.105
T460 18973-19007 Sentence denotes Interpretation of the above table.
T461 19008-19118 Sentence denotes Each row corresponds to an association score whereas each column corresponds to one of the evaluation metrics.
T462 19119-19264 Sentence denotes A (+) in the column means a higher evaluation metric value, the better the association score in that row separates the positive and random pairs.
T463 19265-19313 Sentence denotes A (-) means a lower evaluation metric is better.
T464 19314-19415 Sentence denotes Note all the metrics are immune to linear rescalings; also the Mann-Whitney U score is nonparametric.
T465 19416-19754 Sentence denotes While counts matrices have been generated using different technologies (e.g. Drop-Seq, 10x Genomics, etc.) and different alignment/pre-processing pipelines, all counts matrices were scaled such that each cell contains a total of 10,000 scaled counts (i.e. the sum of expression values for all genes equals 10,000 in each individual cell).
T466 19755-19839 Sentence denotes All data were uniformly processed using the Seurat v3 package (Butler et al., 2018).
T467 19840-19893 Sentence denotes In short, this pipeline involves the following steps.
T468 19894-20045 Sentence denotes First, we identify 2000 variable genes across the given dataset and then perform linear dimensionality reduction by principal component analysis (PCA).
T469 20046-20409 Sentence denotes Using the set of principal components which contribute >80% of variance across the dataset, we then do the following: (i) perform graph-based clustering to identify groups of cells with similar expression profiles (Louvain clustering), (ii) compute UMAP and tSNE coordinates for each individual cell (used for data visualization) and (iii) annotate cell clusters.
T470 20410-20665 Sentence denotes Note that the three human pancreatic datasets (GSE81076, GSE85241, GSE86469) were integrated together in a shared multi-dimensional space using CCA (Canonical Correlation Analysis) and the integration method in the Seurat v3 package (Butler et al., 2018).
T471 20666-20780 Sentence denotes Cell clustering and computation of dimensionality reduction coordinates were performed on this integrated dataset.
T472 20782-20805 Sentence denotes Cell cluster annotation
T473 20806-21031 Sentence denotes In cases where publicly deposited counts matrices are accompanied by author-assigned annotations for individual cells or clusters, we have retained these cell annotations for display in the platform and accompanying analyses.
T474 21032-21416 Sentence denotes For any study which was not accompanied by a metadata file containing cluster annotations, we have manually labeled clusters based on sets of canonical ‘cluster-defining genes.’ In these cases, we have attempted to leverage annotations and descriptions of gene expression patterns described by study authors in the manuscript text and figures corresponding to the data being analyzed.
T475 21418-21468 Sentence denotes Metrics to summarize cluster-level gene expression
T476 21469-21535 Sentence denotes The platform allows users to query any gene in any selected study.
T477 21536-21683 Sentence denotes The corresponding data is displayed in commonly employed formats including a series of violin plots and as a set of dimensionality reduction plots.
T478 21684-21835 Sentence denotes Expression is summarized by listing the percent of cells expressing Gene G in each annotated cluster and the mean expression of Gene G in each cluster.
T479 21836-22070 Sentence denotes To measure the specificity of Gene G expression to each Cluster C, we compute a Cohen’s D value which assesses the effect size between the mean expression of Gene G in cluster C and the mean expression of Gene G in all other clusters.
T480 22071-22283 Sentence denotes Specifically, the Cohen’s D formula is given as follows: (MeanC - MeanA)/(sqrt(StDevC2 + StDevA2)) , where C represents the cluster of interest and A represents the complement of C (i.e. all other cell clusters).
T481 22284-22461 Sentence denotes Note that this is functionally similar to the computation of paired fold change values and p-values between clusters which is frequently used to identify cluster-defining genes.
T482 22463-22490 Sentence denotes Gene-gene cosine similarity
T483 22491-22641 Sentence denotes Within the platform, we support the run-time computation of cosine similarity (i.e. 1 - cosine distance) between the queried gene and all other genes.
T484 22642-22768 Sentence denotes This provides a measure of expression similarity across cells and can be used to identify co-regulated and co-expressed genes.
T485 22769-22868 Sentence denotes Specifically, to perform this computation, we construct a ‘gene expression vector’ for each gene G.
T486 22869-23000 Sentence denotes This corresponds to the set of CP10K values for gene G in each individual cell from the selected populations in the selected study.
T487 23002-23071 Sentence denotes Profiling expression of coronavirus receptors in single-cell datasets
T488 23072-23163 Sentence denotes For each single-cell dataset, we examined the expression of ACE2, TMPRSS2, ANPEP, and DPP4.
T489 23164-23318 Sentence denotes We generally considered a cell population to potentially express a gene if at least 5% of cells from that cluster showed non-zero expression of this gene.
T490 23319-23549 Sentence denotes For each dataset, we show a figure which includes a UMAP dimensionality reduction plot colored by annotated cell type along with identical plots colored by the expression level of each coronavirus receptor in all individual cells.
T491 23550-23805 Sentence denotes In some cases, we also show violin plots from the platform which automatically integrate literature-derived insights to highlight whether there exist textual associations between the queried gene and the tissue/cell types identified in the selected study.
T492 23807-23858 Sentence denotes FDA Adverse Event Reporting System (FAERS) analysis
T493 23859-24075 Sentence denotes The FAERS application of the nferX platform supports viewing adverse event profiles of all marketed products through multiple lenses - Count, Proportional Reporting Ratio (PRR), and an nferX Adverse Event (AE) Score.
T494 24076-24184 Sentence denotes AEScore=ln(count)∗1/(1+e−(prr−1.5)). Count is the raw number of reports between a drug and an adverse event.
T495 24185-24373 Sentence denotes The proportional reporting ratio (PRR) is a simple way to get a measure of how common an adverse event for a particular drug is compared to how common the event is in the overall database.
T496 24374-24732 Sentence denotes A PRR >1 for a drug-event combination indicates that a greater proportion of the reports for the drug are for the event than the proportion of events in the rest of the database, while a PRR of 2 for a drug event combination indicates that the proportion of reports for the drug-event combination is twice the proportion of the event in the overall database.
T497 24733-24787 Sentence denotes The PRR is computed as follows:PRR=(m/n)/((M−m)/(N−n))
T498 24788-24829 Sentence denotes m = number of reports with drug and event
T499 24830-24861 Sentence denotes n = number of reports with drug
T500 24862-24906 Sentence denotes M = number of reports with event in database
T501 24907-24940 Sentence denotes N = number of reports in database
T502 24941-25016 Sentence denotes Count of an event with a query drug is a good first measure of association.
T503 25017-25217 Sentence denotes But it has the problem that generally common events will often show up at the top, where we are often more interested in events that are differentially associated with the query drug over other drugs.
T504 25218-25304 Sentence denotes An issue with PRR is that it is noisy when the total number of event reports is small.
T505 25305-25701 Sentence denotes If there are three reports of some oddly specific event and one occurs with the query drug, that event will likely have a very high PRR, but it may not be the event we would be most interested in for a drug (in FAERS such rare events are often not even proper adverse events) - we want events that occur often, and also are differentially associated with a drug - a balance between count and PRR.
T506 25702-25769 Sentence denotes The AE score tries to strike this balance in an all-in-one measure.
T507 25770-25943 Sentence denotes It up-weights events that occur often for the query drug (this is the ln(count) term), and that are differentially associated with the query drug (this is the sigmoid term).
T508 25944-25998 Sentence denotes The sigmoid(PRR-1.5) term ranges smoothly from 0 to 1.
T509 25999-26030 Sentence denotes It's equal to 0.5 at PRR = 1.5.
T510 26031-26140 Sentence denotes When PRR = 6, sigmoid(PRR-1.5)=0.99; so PRR values >= 6 are all treated roughly equivalently by the AE score.
T511 26141-26361 Sentence denotes Thus, extremely high PRRs due to small counts will not swing the AE score much beyond PRR = 6, and the ln(count) term will down-weight those small-count cases, so that they do not show up at the top of the AE score list.
T512 26362-26548 Sentence denotes A nice property of AE score is that, for a given query drug, the AE scores of the events with that drug turn out to roughly follow an exponential distribution, particularly at the tails.
T513 26549-26623 Sentence denotes We can then fit exponential distributions to the scores, and analyze them.
T514 26624-26833 Sentence denotes A benefit of the exponential fit is that we can make more robust claims about how significant a certain score is for a query drug, even if the empirical data is sparse/noisy at the tails for a particular drug.