AI Psychometrics: Assessing the Psychological Profiles of Large Language Models Through Psychometric Inventories.

We illustrate how standard psychometric inventories originally designed for assessing noncognitive human traits can be repurposed as diagnostic tools to evaluate analogous traits in large language models (LLMs). We start from the assumption that LLMs, inadvertently yet inevitably, acquire psychological traits (metaphorically speaking) from the vast text corpora on which they are trained. Such corpora contain sediments of the personalities, values, beliefs, and biases of the countless human authors of these texts, which LLMs learn through a complex training process. The traits that LLMs acquire in such a way can potentially influence their behavior, that is, their outputs in downstream tasks and applications in which they are employed, which in turn may have real-world consequences for individuals and social groups. By eliciting LLMs' responses to language-based psychometric inventories, we can bring their traits to light. Psychometric profiling enables researchers to study and compare LLMs in terms of noncognitive characteristics, thereby providing a window into the personalities, values, beliefs, and biases these models exhibit (or mimic). We discuss the history of similar ideas and outline possible psychometric approaches for LLMs. We demonstrate one promising approach, zero-shot classification, for several LLMs and psychometric inventories. We conclude by highlighting open challenges and future avenues of research for AI Psychometrics.

Feasibility and Safety of Personalized, Multi-Target, Adoptive Cell Therapy (IMA101): First-in-Human Clinical Trial in Patients with Advanced Metastatic Cancer.

IMA101 is an actively personalized, multi-targeted adoptive cell therapy (ACT), whereby autologous T cells are directed against multiple novel defined peptide-HLA (pHLA) cancer targets. HLA-A*02:01-positive patients with relapsed/refractory solid tumors expressing ≥1 of 8 predefined targets underwent leukapheresis. Endogenous T cells specific for up to 4 targets were primed and expanded in vitro. Patients received lymphodepletion (fludarabine, cyclophosphamide), followed by T-cell infusion and low-dose IL2 (Cohort 1). Patients in Cohort 2 received atezolizumab for up to 1 year (NCT02876510). Overall, 214 patients were screened, 15 received lymphodepletion (13 women, 2 men; median age, 44 years), and 14 were treated with T-cell products. IMA101 treatment was feasible and well tolerated. The most common adverse events were cytokine release syndrome (Grade 1, n = 6; Grade 2, n = 4) and expected cytopenias. No patient died during the first 100 days after T-cell therapy. No neurotoxicity was observed. No objective responses were noted. Prolonged disease stabilization was noted in three patients lasting for 13.7, 12.9, and 7.3 months. High frequencies of target-specific T cells (up to 78.7% of CD8+ cells) were detected in the blood of treated patients, persisted for >1 year, and were detectable in posttreatment tumor tissue. Individual T-cell receptors (TCR) contained in T-cell products exhibited broad variation in TCR avidity, with the majority being low avidity. High-avidity TCRs were identified in some patients' products. This study demonstrates the feasibility and tolerability of an actively personalized ACT directed to multiple defined pHLA cancer targets. Results warrant further evaluation of multi-target ACT approaches using potent high-avidity TCRs. See related Spotlight by Uslu and June, p. 865.

An ultra-high-throughput screen for the evaluation of peptide HLA-Binder interactions.

Peptide human leukocyte antigen (pHLA) targeting therapeutics like T-cell receptor based adoptive cell therapy or bispecific T cell engaging receptor molecules hold great promise for the treatment of cancer. Comprehensive pre-clinical screening of therapeutic candidates is important to ensure patient safety but is challenging because of the size of the potential off-target space. By combining stabilized peptide-receptive HLA molecules with microarray printing and screening, we have developed an ultra-high-throughput screening platform named ValidaTe that enables large scale evaluation of pHLA-binder interactions. We demonstrate its potential by measuring and analyzing over 30.000 binding curves for a high-affinity T cell Engaging Receptor towards a large pHLA library. Compared to a dataset obtained by conventional bio-layer interferometry measurements, we illustrate that a massively increased throughput (over 650 fold) is obtained by our microarray screening, paving the way for use in pre-clinical safety screening of pHLA-targeting drugs.

Inequality and fairness with heterogeneous endowments.

People differ in intelligence, cognitive ability, personality traits, motivation, and similar valued and, to a large degree, inherited characteristics that determine success and achievements. When does individual heterogeneity lead to a fair distribution of rewards and outcomes? Here, we develop this question theoretically and then test it experimentally for a set of structural conditions in a specific interaction situation. We first catalogue the functional relationship between individual endowments and outcomes to distinguish between fairness concepts such as meritocracy, equality of opportunity, equality of outcomes, and Rawl's theory of justice. We then use an online experiment to study which of these fairness patterns emerge when differently endowed individuals can share their resources with others, depending on whether information about others' endowments and outcomes is available. We find that while visible outcomes lessen inequality by decreasing the statistical dispersion of outcomes across the group, endowments need to be visible for better equality of opportunity for the most disadvantaged.

Quantitative immunopeptidomics reveals a tumor stroma-specific target for T cell therapy.

T cell receptor (TCR)-based immunotherapy has emerged as a promising therapeutic approach for the treatment of patients with solid cancers. Identifying peptide-human leukocyte antigen (pHLA) complexes highly presented on tumors and rarely expressed on healthy tissue in combination with high-affinity TCRs that when introduced into T cells can redirect T cells to eliminate tumor but not healthy tissue is a key requirement for safe and efficacious TCR-based therapies. To discover promising shared tumor antigens that could be targeted via TCR-based adoptive T cell therapy, we employed population-scale immunopeptidomics using quantitative mass spectrometry across ~1500 tumor and normal tissue samples. We identified an HLA-A*02:01-restricted pan-cancer epitope within the collagen type VI α-3 (COL6A3) gene that is highly presented on tumor stroma across multiple solid cancers due to a tumor-specific alternative splicing event that rarely occurs outside the tumor microenvironment. T cells expressing natural COL6A3-specific TCRs demonstrated only modest activity against cells presenting high copy numbers of COL6A3 pHLAs. One of these TCRs was affinity-enhanced, enabling transduced T cells to specifically eliminate tumors in vivo that expressed similar copy numbers of pHLAs as primary tumor specimens. The enhanced TCR variants exhibited a favorable safety profile with no detectable off-target reactivity, paving the way to initiate clinical trials using COL6A3-specific TCRs to target an array of solid tumors.

Inequality and inequity in network-based ranking and recommendation algorithms.

Though algorithms promise many benefits including efficiency, objectivity and accuracy, they may also introduce or amplify biases. Here we study two well-known algorithms, namely PageRank and Who-to-Follow (WTF), and show to what extent their ranks produce inequality and inequity when applied to directed social networks. To this end, we propose a directed network model with preferential attachment and homophily (DPAH) and demonstrate the influence of network structure on the rank distributions of these algorithms. Our main findings suggest that (i) inequality is positively correlated with inequity, (ii) inequality is driven by the interplay between preferential attachment, homophily, node activity and edge density, and (iii) inequity is driven by the interplay between homophily and minority size. In particular, these two algorithms reduce, replicate and amplify the representation of minorities in top ranks when majorities are homophilic, neutral and heterophilic, respectively. Moreover, when this representation is reduced, minorities may improve their visibility in the rank by connecting strategically in the network. For instance, by increasing their out-degree or homophily when majorities are also homophilic. These findings shed light on the social and algorithmic mechanisms that hinder equality and equity in network-based ranking and recommendation algorithms.

SARS-CoV-2-reactive T-cell receptors isolated from convalescent COVID-19 patients confer potent T-cell effector function.

Both B cells and T cells are involved in an effective immune response to SARS-CoV-2, the disease-causing virus of COVID-19. While B cells-with the indispensable help of CD4+ T cells-are essential to generate neutralizing antibodies, T cells on their own have been recognized as another major player in effective anti-SARS-CoV-2 immunity. In this report, we provide insights into the characteristics of individual HLA-A*02:01- and HLA-A*24:02-restricted SARS-CoV-2-reactive TCRs, isolated from convalescent COVID-19 patients. We observed that SARS-CoV-2-reactive T-cell populations were clearly detectable in convalescent samples and that TCRs isolated from these T cell clones were highly functional upon ectopic re-expression. The SARS-CoV-2-reactive TCRs described in this report mediated potent TCR signaling in reporter assays with low nanomolar EC50 values. We further demonstrate that these SARS-CoV-2-reactive TCRs conferred powerful T-cell effector function to primary CD8+ T cells as evident by a robust anti-SARS-CoV-2 IFN-γ response and in vitro cytotoxicity. We also provide an example of a long-lasting anti-SARS-CoV-2 memory response by reisolation of one of the retrieved TCRs 5 months after initial sampling. Taken together, these findings contribute to a better understanding of anti-SARS-CoV-2 T-cell immunity and may contribute to paving the way toward immunotherapeutics approaches targeting SARS-CoV-2.

Measuring algorithmically infused societies.

It has been the historic responsibility of the social sciences to investigate human societies. Fulfilling this responsibility requires social theories, measurement models and social data. Most existing theories and measurement models in the social sciences were not developed with the deep societal reach of algorithms in mind. The emergence of 'algorithmically infused societies'-societies whose very fabric is co-shaped by algorithmic and human behaviour-raises three key challenges: the insufficient quality of measurements, the complex consequences of (mis)measurements, and the limits of existing social theories. Here we argue that tackling these challenges requires new social theories that account for the impact of algorithmic systems on social realities. To develop such theories, we need new methodologies for integrating data and measurements into theory construction. Given the scale at which measurements can be applied, we believe measurement models should be trustworthy, auditable and just. To achieve this, the development of measurements should be transparent and participatory, and include mechanisms to ensure measurement quality and identify possible harms. We argue that computational social scientists should rethink what aspects of algorithmically infused societies should be measured, how they should be measured, and the consequences of doing so.

Homophily and minority-group size explain perception biases in social networks.

People's perceptions about the size of minority groups in social networks can be biased, often showing systematic over- or underestimation. These social perception biases are often attributed to biased cognitive or motivational processes. Here we show that both over- and underestimation of the size of a minority group can emerge solely from structural properties of social networks. Using a generative network model, we show that these biases depend on the level of homophily, its asymmetric nature and on the size of the minority group. Our model predictions correspond well with empirical data from a cross-cultural survey and with numerical calculations from six real-world networks. We also identify circumstances under which individuals can reduce their biases by relying on perceptions of their neighbours. This work advances our understanding of the impact of network structure on social perception biases and offers a quantitative approach for addressing related issues in society.

High-throughput peptide-MHC complex generation and kinetic screenings of TCRs with peptide-receptive HLA-A*02:01 molecules.

Major histocompatibility complex (MHC) class I molecules present short peptide ligands on the cell surface for interrogation by cytotoxic CD8+ T cells. MHC class I complexes presenting tumor-associated peptides such as neoantigens represent key targets of cancer immunotherapy approaches currently in development, making them important for efficacy and safety screenings. Without peptide ligand, MHC class I complexes are unstable and decay quickly, making the production of soluble monomers for analytical purposes labor intensive. We have developed a disulfide-stabilized HLA-A*02:01 molecule that is stable without peptide but can form peptide-MHC complexes (pMHCs) with ligands of choice in a one-step loading procedure. We illustrate the similarity between the engineered mutant and the wild-type molecule with respect to affinity of wild-type or affinity-matured T cell receptors (TCRs) and present a crystal structure corroborating the binding kinetics measurements. In addition, we demonstrate a high-throughput binding kinetics measurement platform to analyze the binding characteristics of bispecific TCR (bsTCR) molecules against diverse pMHC libraries produced with the disulfide-stabilized HLA-A*02:01 molecule. We show that bsTCR affinities for pMHCs are indicative of in vitro function and generate a bsTCR binding motif to identify potential off-target interactions in the human proteome. These findings showcase the potential of the platform and the engineered HLA-A*02:01 molecule in the emerging field of pMHC-targeting biologics.

Secretory Acid Sphingomyelinase in the Serum of Medicated Patients Predicts the Prospective Course of Depression.

Major depressive disorder (MDD) is a highly prevalent and devastating psychiatric illness with strong individual and societal burdens. However, biomarkers to improve the limited preventive and therapeutic approaches are scarce. Multilevel evidence suggests that the pathophysiological involvement of sphingolipids particularly increases the levels of ceramides and the ceramide hydrolyzing enzyme, acid sphingomyelinase. The activity of secretory acid sphingomyelinase (S-ASM) and routine blood parameters were determined in the serum of patients with current (unmedicated n = 63, medicated n = 66) and remitted (n = 39) MDD and healthy subjects (n = 61). Depression severity and anxiety and their 3-weeks prospective course of treatment were assessed by psychometric inventories. S-ASM activity was not different between the four groups, did not decrease during treatment, and was not lower in individuals taking medication that functionally inhibited ASM. However, S-ASM correlated positively with depression severity only in remitted patients. High enzyme activity at inclusion predicted milder clinician-evaluated and self-rated depression severity (HAM-D, MADRS, BDI-II) and state anxiety at follow-up, and was related to stronger improvement in these scores in medicated patients. S-ASM was strongly and contrariwise associated with serum lipids in unmedicated and medicated females. These findings contribute to a better understanding of the pathomechanisms underlying depression and the development of clinical strategies and biomarkers.

LDL cholesterol relates to depression, its severity, and the prospective course.

BACKGROUND: Major depressive disorder (MDD) is a highly prevalent and burdening mental illness. Approximately 30% of the major depressive episodes (MDE) are classified as therapy-refractory. Further knowledge of the pathophysiological mechanisms underlying MDD and predictive biomarkers are needed to improve treatment options. METHODS: Serum lipid levels were compared between patients with a current MDE (n = 130) or remitted MDD (n = 39) and healthy control subjects (n = 61) and associated with the severity (17-item Hamilton Depression Rating Scale [HAMD] scores) and the prospective course of depression (direct follow-up of at median 20 days post-inclusion). RESULTS: We found higher levels of LDL cholesterol (152.5 vs. 134.0 mg/dl, U = 3021, P = 0.008) and LDL/HDL ratio (2.82 vs. 2.21, U = 2912, P = 0.003) in patients with a current MDE than in healthy control subjects. In patients with a current MDE, higher HAMD scores correlated also with higher values of triglycerides (ρ = 0.213, P = 0.015), total cholesterol (ρ = 0.199, P = 0.023), LDL cholesterol (ρ = 0.224, P = 0.010), and LDL/HDL ratio (ρ = 0.196, P = 0.026). Moreover, higher total cholesterol (ρ = -0.233, P = 0.010), LDL cholesterol (ρ = -0.235, P = 0.010), and LDL/HDL ratio (ρ = -0.199, P = 0.029) were associated with a stronger decline in HAMD score between study inclusion and direct follow-up. LIMITATIONS: We employed an associational study design, performed only a short-term follow-up, and excluded suicidal study subjects. CONCLUSIONS: Serum lipid levels are associated with depression per se, the depression severity, and the prospective 3-week course. These observations build the basis for future investigations on individualized lipid metabolism-related treatment strategies in depressed patients.

Publisher Correction: Actively personalized vaccination trial for newly diagnosed glioblastoma.

The additional author support information was erroneously omitted from the Supplementary Information. This has been corrected online.

Actively personalized vaccination trial for newly diagnosed glioblastoma.

Patients with glioblastoma currently do not sufficiently benefit from recent breakthroughs in cancer treatment that use checkpoint inhibitors1,2. For treatments using checkpoint inhibitors to be successful, a high mutational load and responses to neoepitopes are thought to be essential3. There is limited intratumoural infiltration of immune cells4 in glioblastoma and these tumours contain only 30-50 non-synonymous mutations5. Exploitation of the full repertoire of tumour antigens-that is, both unmutated antigens and neoepitopes-may offer more effective immunotherapies, especially for tumours with a low mutational load. Here, in the phase I trial GAPVAC-101 of the Glioma Actively Personalized Vaccine Consortium (GAPVAC), we integrated highly individualized vaccinations with both types of tumour antigens into standard care to optimally exploit the limited target space for patients with newly diagnosed glioblastoma. Fifteen patients with glioblastomas positive for human leukocyte antigen (HLA)-A*02:01 or HLA-A*24:02 were treated with a vaccine (APVAC1) derived from a premanufactured library of unmutated antigens followed by treatment with APVAC2, which preferentially targeted neoepitopes. Personalization was based on mutations and analyses of the transcriptomes and immunopeptidomes of the individual tumours. The GAPVAC approach was feasible and vaccines that had poly-ICLC (polyriboinosinic-polyribocytidylic acid-poly-L-lysine carboxymethylcellulose) and granulocyte-macrophage colony-stimulating factor as adjuvants displayed favourable safety and strong immunogenicity. Unmutated APVAC1 antigens elicited sustained responses of central memory CD8+ T cells. APVAC2 induced predominantly CD4+ T cell responses of T helper 1 type against predicted neoepitopes.

Homophily influences ranking of minorities in social networks.

Homophily can put minority groups at a disadvantage by restricting their ability to establish links with a majority group or to access novel information. Here, we show how this phenomenon can influence the ranking of minorities in examples of real-world networks with various levels of heterophily and homophily ranging from sexual contacts, dating contacts, scientific collaborations, and scientific citations. We devise a social network model with tunable homophily and group sizes, and demonstrate how the degree ranking of nodes from the minority group in a network is a function of (i) relative group sizes and (ii) the presence or absence of homophilic behaviour. We provide analytical insights on how the ranking of the minority can be improved to ensure the representativeness of the group and correct for potential biases. Our work presents a foundation for assessing the impact of homophilic and heterophilic behaviour on minorities in social networks.

The emergence of inequality in social groups: Network structure and institutions affect the distribution of earnings in cooperation games.

From small communities to entire nations and society at large, inequality in wealth, social status, and power is one of the most pervasive and tenacious features of the social world. What causes inequality to emerge and persist? In this study, we investigate how the structure and rules of our interactions can increase inequality in social groups. Specifically, we look into the effects of four structural conditions-network structure, network fluidity, reputation tracking, and punishment institutions-on the distribution of earnings in network cooperation games. We analyze 33 experiments comprising 96 experimental conditions altogether. We find that there is more inequality in clustered networks compared to random networks, in fixed networks compared to randomly rewired and strategically updated networks, and in groups with punishment institutions compared to groups without. Secondary analyses suggest that the reasons inequality emerges under these conditions may have to do with the fact that fixed networks allow exploitation of the poor by the wealthy and clustered networks foster segregation between the poor and the wealthy, while the burden of costly punishment falls onto the poor, leaving them poorer. Surprisingly, we do not find evidence that inequality is affected by reputation in a systematic way but this could be because reputation needs to play out in a particular network environment in order to have an effect. Overall, our findings suggest possible strategies and interventions to decrease inequality and mitigate its negative impact, particularly in the context of mid- and large-sized organizations and online communities.

Micro-organisms growing on rapeseed during storage affect the profile of volatile compounds of virgin rapeseed oil.

BACKGROUND: Micro-organisms populate on rapeseed after harvest during storage depending on the growing conditions. The composition of the bacterial colonization is unknown, although its contribution to the profile of volatile aroma-active compounds determines the sensory quality of virgin cold-pressed rapeseed oil. RESULTS: From four rapeseed samples, 46 bacterial strains were isolated. By DNA-sequencing, the identification of four bacteria species and 17 bacteria genera was possible. In total, 22 strains were selected, based on their typical off-flavors resembling those of virgin sensory bad cold-pressed rapeseed oils. The cultivation of these strains on rapeseed meal agar and examination of volatile compounds by solid phase microextraction-gas chromatography-mass spectrometry allowed the identification of 29 different compounds, mainly degradation products of fatty acids such as alkanes, alkenes, aldehydes, ketones and alcohols and, in addition, sulfur-containing compounds, including one terpene and three pyrazines. From these compounds, 19 are described as aroma-active in the literature. CONCLUSION: Micro-organisms populating on rapeseed during storage may strongly influence the sensory quality of virgin rapeseed oil as a result of the development of volatile aroma-active metabolic products. It can be assumed that occurrence of off-flavor of virgin rapeseed oils on the market are the result of metabolic degradation products produced by micro-organisms populating on rapeseed during storage. © 2017 Society of Chemical Industry.