Center for Information Technology

Politics and political conflict often occur in the written and spoken word. Scholars have long recognized this, but the massive costs of analyzing even moderately sized collections of texts have hindered their use in political science research. Here lies the promise of automated text analysis: it substantially reduces the costs of analyzing large collections of text. We provide a guide to this exciting new area of research and show how, in many instances, the methods have already obtained part of their promise. But there are pitfalls to using automated methods—they are no substitute for careful thought and close reading and require extensive and problem-specific validation. We survey a wide range of new methods, provide guidance on how to validate the output of the models, and clarify misconceptions and errors in the literature. To conclude, we argue that for automated text methods to become a standard tool for political scientists, methodologists must contribute new methods and new methods of validation.

Fiber tract trajectories in coherently organized brain white matter pathways were computed from in vivo diffusion tensor magnetic resonance imaging (DT-MRI) data. First, a continuous diffusion tensor field is constructed from this discrete, noisy, measured DT-MRI data. Then a Frenet equation, describing the evolution of a fiber tract, was solved. This approach was validated using synthesized, noisy DT-MRI data. Corpus callosum and pyramidal tract trajectories were constructed and found to be consistent with known anatomy. The method's reliability, however, degrades where the distribution of fiber tract directions is nonuniform. Moreover, background noise in diffusion-weighted MRIs can cause a computed trajectory to hop from tract to tract. Still, this method can provide quantitative information with which to visualize and study connectivity and continuity of neural pathways in the central and peripheral nervous systems in vivo, and holds promise for elucidating architectural features in other fibrous tissues and ordered media.

Machine learning is a means to derive artificial intelligence by discovering patterns in existing data. Here, we show that applying machine learning to ordinary human language results in human-like semantic biases. We replicated a spectrum of known biases, as measured by the Implicit Association Test, using a widely used, purely statistical machine-learning model trained on a standard corpus of text from the World Wide Web. Our results indicate that text corpora contain recoverable and accurate imprints of our historic biases, whether morally neutral as toward insects or flowers, problematic as toward race or gender, or even simply veridical, reflecting the status quo distribution of gender with respect to careers or first names. Our methods hold promise for identifying and addressing sources of bias in culture, including technology.

Recursive partitioning methods have become popular and widely used tools for nonparametric regression and classification in many scientific fields. Especially random forests, which can deal with large numbers of predictor variables even in the presence of complex interactions, have been applied successfully in genetics, clinical medicine, and bioinformatics within the past few years. High-dimensional problems are common not only in genetics, but also in some areas of psychological research, where only a few subjects can be measured because of time or cost constraints, yet a large amount of data is generated for each subject. Random forests have been shown to achieve a high prediction accuracy in such applications and to provide descriptive variable importance measures reflecting the impact of each variable in both main effects and interactions. The aim of this work is to introduce the principles of the standard recursive partitioning methods as well as recent methodological improvements, to illustrate their usage for low and high-dimensional data exploration, but also to point out limitations of the methods and potential pitfalls in their practical application. Application of the methods is illustrated with freely available implementations in the R system for statistical computing.

A systematic analysis of the RNA splice junction sequences of eukaryotic protein coding genes was carried out using the GENBANK databank. Nucleotide frequencies obtained for the highly conserved regions around the splice sites for different categories of organisms closely agree with each other. A striking similarity among the rare splice junctions which do not contain AG at the 3' splice site or GT at the 5' splice site indicates the existence of special mechanisms to recognize them, and that these unique signals may be involved in crucial gene-regulation events and in differentiation. A method was developed to predict potential exons in a bare sequence, using a scoring and ranking scheme based on nucleotide weight tables. This method was used to find a majority of the exons in selected known genes, and also predicted potential new exons which may be used in alternative splicing situations.

BACKGROUND: Diffuse large B-cell lymphomas (DLBCLs) are phenotypically and genetically heterogeneous. Gene-expression profiling has identified subgroups of DLBCL (activated B-cell-like [ABC], germinal-center B-cell-like [GCB], and unclassified) according to cell of origin that are associated with a differential response to chemotherapy and targeted agents. We sought to extend these findings by identifying genetic subtypes of DLBCL based on shared genomic abnormalities and to uncover therapeutic vulnerabilities based on tumor genetics. METHODS: We studied 574 DLBCL biopsy samples using exome and transcriptome sequencing, array-based DNA copy-number analysis, and targeted amplicon resequencing of 372 genes to identify genes with recurrent aberrations. We developed and implemented an algorithm to discover genetic subtypes based on the co-occurrence of genetic alterations. RESULTS: and CD79B mutations), BN2 (based on BCL6 fusions and NOTCH2 mutations), N1 (based on NOTCH1 mutations), and EZB (based on EZH2 mutations and BCL2 translocations). Genetic aberrations in multiple genes distinguished each genetic subtype from other DLBCLs. These subtypes differed phenotypically, as judged by differences in gene-expression signatures and responses to immunochemotherapy, with favorable survival in the BN2 and EZB subtypes and inferior outcomes in the MCD and N1 subtypes. Analysis of genetic pathways suggested that MCD and BN2 DLBCLs rely on "chronic active" B-cell receptor signaling that is amenable to therapeutic inhibition. CONCLUSIONS: We uncovered genetic subtypes of DLBCL with distinct genotypic, epigenetic, and clinical characteristics, providing a potential nosology for precision-medicine strategies in DLBCL. (Funded by the Intramural Research Program of the National Institutes of Health and others.).

BACKGROUND: Despite best current therapy, up to 20% of pediatric patients with acute lymphoblastic leukemia (ALL) have a relapse. Recent genomewide analyses have identified a high frequency of DNA copy-number abnormalities in ALL, but the prognostic implications of these abnormalities have not been defined. METHODS: We studied a cohort of 221 children with high-risk B-cell-progenitor ALL with the use of single-nucleotide-polymorphism microarrays, transcriptional profiling, and resequencing of samples obtained at diagnosis. Children with known very-high-risk ALL subtypes (i.e., BCR-ABL1-positive ALL, hypodiploid ALL, and ALL in infants) were excluded from this cohort. A copy-number abnormality was identified as a predictor of poor outcome, and it was then tested in an independent validation cohort of 258 patients with B-cell-progenitor ALL. RESULTS: More than 50 recurring copy-number abnormalities were identified, most commonly involving genes that encode regulators of B-cell development (in 66.8% of patients in the original cohort); PAX5 was involved in 31.7% and IKZF1 in 28.6% of patients. Using copy-number abnormalities, we identified a predictor of poor outcome that was validated in the independent validation cohort. This predictor was strongly associated with alteration of IKZF1, a gene that encodes the lymphoid transcription factor IKAROS. The gene-expression signature of the group of patients with a poor outcome revealed increased expression of hematopoietic stem-cell genes and reduced expression of B-cell-lineage genes, and it was similar to the signature of BCR-ABL1-positive ALL, another high-risk subtype of ALL with a high frequency of IKZF1 deletion. CONCLUSIONS: Genetic alteration of IKZF1 is associated with a very poor outcome in B-cell-progenitor ALL.

BACKGROUND: Patients with follicular lymphoma may survive for periods of less than 1 year to more than 20 years after diagnosis. We used gene-expression profiles of tumor-biopsy specimens obtained at diagnosis to develop a molecular predictor of the length of survival. METHODS: Gene-expression profiling was performed on 191 biopsy specimens obtained from patients with untreated follicular lymphoma. Supervised methods were used to discover expression patterns associated with the length of survival in a training set of 95 specimens. A molecular predictor of survival was constructed from these genes and validated in an independent test set of 96 specimens. RESULTS: Individual genes that predicted the length of survival were grouped into gene-expression signatures on the basis of their expression in the training set, and two such signatures were used to construct a survival predictor. The two signatures allowed patients with specimens in the test set to be divided into four quartiles with widely disparate median lengths of survival (13.6, 11.1, 10.8, and 3.9 years), independently of clinical prognostic variables. Flow cytometry showed that these signatures reflected gene expression by nonmalignant tumor-infiltrating immune cells. CONCLUSIONS: The length of survival among patients with follicular lymphoma correlates with the molecular features of nonmalignant immune cells present in the tumor at diagnosis.

BACKGROUND: For the treatment of myocardial infarction with ST-segment elevation, primary angioplasty is considered superior to fibrinolysis for patients who are admitted to hospitals with angioplasty facilities. Whether this benefit is maintained for patients who require transportation from a community hospital to a center where invasive treatment is available is uncertain. METHODS: We randomly assigned 1572 patients with acute myocardial infarction to treatment with angioplasty or accelerated treatment with intravenous alteplase; 1129 patients were enrolled at 24 referral hospitals and 443 patients at 5 invasive-treatment centers. The primary study end point was a composite of death, clinical evidence of reinfarction, or disabling stroke at 30 days. RESULTS: Among patients who underwent randomization at referral hospitals, the primary end point was reached in 8.5 percent of the patients in the angioplasty group, as compared with 14.2 percent of those in the fibrinolysis group (P=0.002). The results were similar among patients who were enrolled at invasive-treatment centers: 6.7 percent of the patients in the angioplasty group reached the primary end point, as compared with 12.3 percent in the fibrinolysis group (P=0.05). Among all patients, the better outcome after angioplasty was driven primarily by a reduction in the rate of reinfarction (1.6 percent in the angioplasty group vs. 6.3 percent in the fibrinolysis group, P<0.001); no significant differences were observed in the rate of death (6.6 percent vs. 7.8 percent, P=0.35) or the rate of stroke (1.1 percent vs. 2.0 percent, P=0.15). Ninety-six percent of patients were transferred from referral hospitals to an invasive-treatment center within two hours. CONCLUSIONS: A strategy for reperfusion involving the transfer of patients to an invasive-treatment center for primary angioplasty is superior to on-site fibrinolysis, provided that the transfer takes two hours or less.

Gene-expression profiling has been used to define 3 molecular subtypes of diffuse large B-cell lymphoma (DLBCL), termed germinal center B-cell-like (GCB) DLBCL, activated B-cell-like (ABC) DLBCL, and primary mediastinal B-cell lymphoma (PMBL). To investigate whether these DLBCL subtypes arise by distinct pathogenetic mechanisms, we analyzed 203 DLBCL biopsy samples by high-resolution, genome-wide copy number analysis coupled with gene-expression profiling. Of 272 recurrent chromosomal aberrations that were associated with gene-expression alterations, 30 were used differentially by the DLBCL subtypes (P < 0.006). An amplicon on chromosome 19 was detected in 26% of ABC DLBCLs but in only 3% of GCB DLBCLs and PMBLs. A highly up-regulated gene in this amplicon was SPIB, which encodes an ETS family transcription factor. Knockdown of SPIB by RNA interference was toxic to ABC DLBCL cell lines but not to GCB DLBCL, PMBL, or myeloma cell lines, strongly implicating SPIB as an oncogene involved in the pathogenesis of ABC DLBCL. Deletion of the INK4a/ARF tumor suppressor locus and trisomy 3 also occurred almost exclusively in ABC DLBCLs and was associated with inferior outcome within this subtype. FOXP1 emerged as a potential oncogene in ABC DLBCL that was up-regulated by trisomy 3 and by more focal high-level amplifications. In GCB DLBCL, amplification of the oncogenic mir-17-92 microRNA cluster and deletion of the tumor suppressor PTEN were recurrent, but these events did not occur in ABC DLBCL. Together, these data provide genetic evidence that the DLBCL subtypes are distinct diseases that use different oncogenic pathways.

BACKGROUND: The distinction between Burkitt's lymphoma and diffuse large-B-cell lymphoma is crucial because these two types of lymphoma require different treatments. We examined whether gene-expression profiling could reliably distinguish Burkitt's lymphoma from diffuse large-B-cell lymphoma. METHODS: Tumor-biopsy specimens from 303 patients with aggressive lymphomas were profiled for gene expression and were also classified according to morphology, immunohistochemistry, and detection of the t(8;14) c-myc translocation. RESULTS: A classifier based on gene expression correctly identified all 25 pathologically verified cases of classic Burkitt's lymphoma. Burkitt's lymphoma was readily distinguished from diffuse large-B-cell lymphoma by the high level of expression of c-myc target genes, the expression of a subgroup of germinal-center B-cell genes, and the low level of expression of major-histocompatibility-complex class I genes and nuclear factor-kappaB target genes. Eight specimens with a pathological diagnosis of diffuse large-B-cell lymphoma had the typical gene-expression profile of Burkitt's lymphoma, suggesting they represent cases of Burkitt's lymphoma that are difficult to diagnose by current methods. Among 28 of the patients with a molecular diagnosis of Burkitt's lymphoma, the overall survival was superior among those who had received intensive chemotherapy regimens instead of lower-dose regimens. CONCLUSIONS: Gene-expression profiling is an accurate, quantitative method for distinguishing Burkitt's lymphoma from diffuse large-B-cell lymphoma.

One of the key enablers of future wireless communications is constituted by massive multiple-input multiple-output (MIMO) systems, which can improve the spectral efficiency by orders of magnitude. In existing massive MIMO systems, however, conventional phased arrays are used for beamforming. This method results in excessive power consumption and high hardware costs. Recently, reconfigurable intelligent surface (RIS) has been considered as one of the revolutionary technologies to enable energy-efficient and smart wireless communications, which is a two-dimensional structure with a large number of passive elements. In this paper, we develop a new type of high-gain yet low-cost RIS that bears 256 elements. The proposed RIS combines the functions of phase shift and radiation together on an electromagnetic surface, where positive intrinsic-negative (PIN) diodes are used to realize 2-bit phase shifting for beamforming. This radical design forms the basis for the world’s first wireless communication prototype using RIS having 256 two-bit elements. The prototype consists of modular hardware and flexible software that encompass the following: the hosts for parameter setting and data exchange, the universal software radio peripherals (USRPs) for baseband and radio frequency (RF) signal processing, as well as the RIS for signal transmission and reception. Our performance evaluation confirms the feasibility and efficiency of RISs in wireless communications. We show that, at 2.3 GHz, the proposed RIS can achieve a 21.7 dBi antenna gain. At the millimeter wave (mmWave) frequency, that is, 28.5 GHz, it attains a 19.1 dBi antenna gain. Furthermore, it has been shown that the RIS-based wireless communication prototype developed is capable of significantly reducing the power consumption.

Abstract-125 years after Bertha Benz completed the first overland journey in automotive history, the Mercedes Benz S-Class S 500 INTELLIGENT DRIVE followed the same route from Mannheim to Pforzheim, Germany, in fully autonomous manner. The autonomous vehicle was equipped with close-to-production sensor hardware and relied solely on vision and radar sensors in combination with accurate digital maps to obtain a comprehensive understanding of complex traffic situations. The historic Bertha Benz Memorial Route is particularly challenging for autonomous driving. The course taken by the autonomous vehicle had a length of 103 km and covered rural roads, 23 small villages and major cities (e.g. downtown Mannheim and Heidelberg). The route posed a large variety of difficult traffic scenarios including intersections with and without traffic lights, roundabouts, and narrow passages with oncoming traffic. This paper gives an overview of the autonomous vehicle and presents details on vision and radar-based perception, digital road maps and video-based self-localization, as well as motion planning in complex urban scenarios.

We present a unified framework for extracting kinetic information from single-molecule pulling experiments at constant force or constant pulling speed. Our procedure provides estimates of not only (i) the intrinsic rate coefficient and (ii) the location of the transition state but also (iii) the free energy of activation. By analyzing simulated data, we show that the resulting rates of force-induced rupture are significantly more reliable than those obtained by the widely used approach based on Bell's formula. We consider the uniqueness of the extracted kinetic information and suggest guidelines to avoid over-interpretation of experiments.

As designing appropriate Convolutional Neural Network (CNN) architecture in the context of a given application usually involves heavy human works or numerous GPU hours, the research community is soliciting the architecture-neutral CNN structures, which can be easily plugged into multiple mature architectures to improve the performance on our real-world applications. We propose Asymmetric Convolution Block (ACB), an architecture-neutral structure as a CNN building block, which uses 1D asymmetric convolutions to strengthen the square convolution kernels. For an off-the-shelf architecture, we replace the standard square-kernel convolutional layers with ACBs to construct an Asymmetric Convolutional Network (ACNet), which can be trained to reach a higher level of accuracy. After training, we equivalently convert the ACNet into the same original architecture, thus requiring no extra computations anymore. We have observed that ACNet can improve the performance of various models on CIFAR and ImageNet by a clear margin. Through further experiments, we attribute the effectiveness of ACB to its capability of enhancing the model's robustness to rotational distortions and strengthening the central skeleton parts of square convolution kernels.

Chronic lymphocytic leukemia (CLL), an incurable malignancy of mature B lymphocytes, involves blood, bone marrow, and secondary lymphoid organs such as the lymph nodes (LN). A role of the tissue microenvironment in the pathogenesis of CLL is hypothesized based on in vitro observations, but its contribution in vivo remains ill-defined. To elucidate the effects of tumor-host interactions in vivo, we purified tumor cells from 24 treatment-naive patients. Samples were obtained concurrently from blood, bone marrow, and/or LN and analyzed by gene expression profiling. We identified the LN as a key site in CLL pathogenesis. CLL cells in the LN showed up-regulation of gene signatures, indicating B-cell receptor (BCR) and nuclear factor-κB activation. Consistent with antigen-dependent BCR signaling and canonical nuclear factor-κB activation, we detected phosphorylation of SYK and IκBα, respectively. Expression of BCR target genes was stronger in clinically more aggressive CLL, indicating more effective BCR signaling in this subtype in vivo. Tumor proliferation, quantified by the expression of the E2F and c-MYC target genes and verified with Ki67 staining by flow cytometry, was highest in the LN and was correlated with clinical disease progression. These data identify the disruption of tumor microenvironment interactions and the inhibition of BCR signaling as promising therapeutic strategies in CLL. This study is registered at http://clinicaltrials.gov as NCT00019370.

OBJECTIVE: Falls are a well-known source of morbidity and mortality in the elderly. Fall-related injury severity in this group, however, is less clear, particularly as it relates to type of fall. Our purpose is to explore the relationship between mechanism of fall and both pattern and severity of injury in geriatric patients as compared with a younger cohort. METHODS: Our trauma registry was queried for all patients evaluated by the trauma service over a 412-year period (1994-1998). Two cohorts were formed on the basis of age greater than 65 or less than or equal to 65 years and compared as to mechanism, Injury Severity Score (ISS), Abbreviated Injury Scale score, and mortality. RESULTS: Over the study period, 1,512 patients were evaluated, 333 greater than 65 years and 1,179 less than or equal to 65 years of age. Falls were the injury mechanism in 48% of the older group and 7% of the younger group (p < 0.05). Falls in the older group constituted 65% of patients with ISS >15, with 32% of all falls resulting in serious injury (ISS >15). In contrast, falls in the younger group constituted only 11% of ISS >15 patients, with falls causing serious injury only 15% of the time (both p < 0.05). Notably, same-level falls resulted in serious injury 30% of the time in the older group versus 4% in the younger group (p < 0.05), and were responsible for an ISS >15 30-fold more in the older group (31% vs. <1%; p < 0.05). Abbreviated Injury Scale evaluation revealed more frequent head/neck (47% vs. 22%), chest (23% vs. 9%), and pelvic/extremity (27% vs. 15%) injuries in the older group for all falls (all p < 0.05). The mean ISS for same-level falls in the older group was twice that for the younger group (9.28 vs. 4.64, p < 0.05), whereas there was no difference in mean ISS between multilevel and same-level falls within the older group itself (10.12 vs. 9.28, p > 0.05). The fall-related death rate was higher in the older group (7% vs. 4%), with falls seven times more likely to be the cause of death compared with the younger group (55% vs. 7.5%) (both p < 0.05). Same-level falls as a cause of death was 10 times more common in the elderly (25% vs. 2.5%, p < 0.05). CONCLUSION: Falls among the elderly, including same-level falls, are a common source of both high injury severity and mortality, much more so than in younger patients. A different pattern of injury between older and younger fall patients also exists.

Machine-learning (ML) methods have gained prominence in the quantitative sciences. However, there are many known methodological pitfalls, including data leakage, in ML-based science. We systematically investigate reproducibility issues in ML-based science. Through a survey of literature in fields that have adopted ML methods, we find 17 fields where leakage has been found, collectively affecting 294 papers and, in some cases, leading to wildly overoptimistic conclusions. Based on our survey, we introduce a detailed taxonomy of eight types of leakage, ranging from textbook errors to open research problems. We propose that researchers test for each type of leakage by filling out model info sheets, which we introduce. Finally, we conduct a reproducibility study of civil war prediction, where complex ML models are believed to vastly outperform traditional statistical models such as logistic regression (LR). When the errors are corrected, complex ML models do not perform substantively better than decades-old LR models.

In our article entitled “Color Schemes to Represent the Orientation of Anisotropic Tissues From Diffusion Tensor Data: Application to White Matter Fiber Tract Mapping in the Human Brain,” which appeared in MRM 42:(3)526–540 Sept., 1999, we incorrectly reported some properties of the color scheme previously proposed by Jones, Williams, and Horsfield (1) (Ref. 26 in our original article). In particular, we incorrectly stated that their scheme would present “discontinuity” artifacts, i.e., sharp artifactual boundaries between complementary colors occurring in structures that are oriented approximately within the plane of the image. As we discuss in our article, these discontinuity artifacts originate from the antipodal symmetry of the eigenvectors of the diffusion tensor, and are present in color maps of fiber direction in which all directions are represented with unique hues (“No symmetry” schemes). “Rotational symmetry” schemes in which vector pairs with azimuthal angles ψ and (ψ + π) are depicted with the same hue do not exhibit these artifacts. After the publication of our article, Jones et al. pointed out to us that their scheme belongs to this second group of color representations and therefore does not suffer from a discontinuity artifact. We apologize to Jones and colleagues for our mistake, which originated from our incorrect interpretation of images they presented at the 1997 meeting of the ISMRM. In their color maps, white matter structures that are known to have a homogeneously varying fiber direction anatomically (such as the corpus callosum) were represented as having sharp color transitions, indicating abrupt changes in fiber orientation. The origin of these sharp transitions, however, should be attributed to perceptual nonuniformities in the color representation scheme they used, rather than to a discontinuity artifact. In the enclosed figure, which replaces Figure 4c in our original publication, Jones' scheme is correctly implemented. As a last remark, we want to underscore that we reported and analyzed Jones' color fiber direction mapping method in our article because it was one of the first that associated color with the components of the eigenvectors of the diffusion tensor. Our misinterpretation of the exact implementation of their method should not diminish the credit that these authors deserve for their interesting and original work: they provided the first example of a tensor-based color scheme which assumes rotational symmetry rather than no-symmetry, as we originally believed.

Disease incidences increase with age, but the molecular characteristics of ageing that lead to increased disease susceptibility remain inadequately understood. Here we perform a whole-blood gene expression meta-analysis in 14,983 individuals of European ancestry (including replication) and identify 1,497 genes that are differentially expressed with chronological age. The age-associated genes do not harbor more age-associated CpG-methylation sites than other genes, but are instead enriched for the presence of potentially functional CpG-methylation sites in enhancer and insulator regions that associate with both chronological age and gene expression levels. We further used the gene expression profiles to calculate the 'transcriptomic age' of an individual, and show that differences between transcriptomic age and chronological age are associated with biological features linked to ageing, such as blood pressure, cholesterol levels, fasting glucose, and body mass index. The transcriptomic prediction model adds biological relevance and complements existing epigenetic prediction models, and can be used by others to calculate transcriptomic age in external cohorts.