Upload your resume. PhdFit's six research agents compare your background with faculty profiles, recent publications, lab focus, and outreach opportunities, then rank professors with evidence you can review.
Resume-aware matchingEvidence from recent researchOutreach drafts included
Advisor Match Brief Resume uploaded
Resume signals5 found
NLPHCIFairnessGraph MLPyTorch
Six-agent analysisRunning
Axel · Candidate AnalystResume parsed
Nova · Professor ResearcherRecent papers
Echo · Match ExplainerEvidence linked
Atlas · Strategy AdvisorFit ranked
Lyra · Outreach WriterDraft queued
Juno · Meeting Prep CompilerNext step ready
SC
Dr. Sarah Chen
Stanford · NLP & interpretability
91
fit
Fairness + NLP projects overlap with her recent work.
Strong methods fit: model evaluation and interpretability.
Clear outreach angle for current trustworthy AI research.
Outreach angle
Ask how her lab is extending interpretability methods into fairness audits for real-world AI systems.
Eric Turkheimer
Verified
University of Virginia · Psychology and Neuroscience
Eric Turkheimer is the Hugh Scott Hamilton Professor of Psychology at the University of Virginia. His research focuses on how interactions between genes and environments shape the development of human behavior. His lab studies various aspects of behavior, with particular interest in issues related to family life, including marriage, divorce, and parenting. Additionally, Turkheimer's work explores the development of human intelligence and personality, especially in understanding the processes that lead siblings to become different from each other over time. He also develops methods for richer and more individualized assessments of personality, controlling for the role of self-esteem when individuals describe their own personality.
Psychiatry is blessed or cursed with theoretical questions that never go away. They mostly take the form of dichotomous oppositions. Nature versus nurture is the most familiar1, but there is also mind versus brain, medication versus talk therapy, conscious versus unconscious, diseases versus problems in living, and categorical versus continuous. These chestnuts have been around so long and debated so many times that by now they are usually dismissed as either settled or irresolvable, yet to everyone’s frustration they refuse to go away. This essay proposes that they are tied together, and could eventually be resolved by consideration of the ways in which behavioral entities are recognized, defined and caused. The universe is organized hierarchically, with big things made out of little ones. Organisms – to which I will limit myself here – are no different. Atoms make molecules, molecules make chemical processes, which make cells, organs, and eventually individuals. Individuals go on to make families, groups and societies. As a hierarchically organized entity changes – that is as it behaves – the changes are manifest at all levels. The tiniest gears turn as a clock tells time, and the hands of the clock move along with motion in the gears. The whole thing churns together. Once again, organisms – now limiting myself further to people – are no different. An individual lapsing into a depressive episode is churning in the cells and genes and neurons, in a marriage and on a therapeutic couch, as part of a history that is personal, cultural and evolutionary. An observer of this multi-level developmental churning is faced with a vastly complex, chaotic system that does not come with ready-made edges to make sense of it. Human observers of human behavior have evolved the ability to create linguistic categories that impose edges on the chaos. Entities created by language regularize developmental chaos by simplifying it. By regularizing it, they make it possible to talk about behavior in systematic ways. Geneticists, neuroscientists, psychiatrists and sociologists create entities at different scales. The parable about the blind men and the elephant misses the point: it is not one scientist holding the trunk and another the tail; it is one with a microscope and another with an aerial camera. Any entity that is created by human observers depends on characteristics that exist at some scale. If we say that an individual is suffering from anaclitic depression, we are referring to edges that are imposed on behavior in personal relationships and in therapy. Observing that a person is a non-responder to selective serotonin reuptake inhibitors (SSRIs) finds edges in pharmacological metabolism and responses on the Beck Depression Inventory. Observing that a person is in a high-risk group based on a polygenic score for depression does not refer to behavior, but rather to the DNA that he/she has inherited. These different entities are to some extent independent of each other, but it is not the case that an entity specified at one scale does not exist at another: it is just less well defined. It becomes blurry2. People diagnosed by their analyst with anaclitic depression may have detectable differences in their SSRI responses or polygenic scores, but psychoanalysts cannot predict drug response, and one cannot identify anaclitic depression in the genes. Except that, sometimes, one can. It occasionally turns out that what appears to be an entity defined at one scale is actually a blurry version of another entity more sharply defined at a different, usually smaller, scale. Chorea is a term from descriptive neurology that refers to a variety of abnormal movements that occur for many different reasons, including phenomena such as hysterical contagion among groups of young people at a school. But in 1872 (i.e., before Mendel), G. Huntington3 noticed that a subset of choreiform movements appeared in families in a structured way, along with a variety of other devastating symptoms. The now eponymous chorea that Huntington called “hereditary” is a large-scale manifestation of a crisply defined entity on a genetic scale. This kind of cross-level reduction is powerful when it occurs, but it must be remembered that it is not inevitable. You could study bankruptcy for the rest of time and never discover a sub-syndrome that is crisply explained by its relation to genetic or neurological processes, despite the fact that bankrupt individuals certainly show blurry genetic and neurological differences from the financially solvent. Behavioral entities are composed of molecules, neurons and genes, but they are caused by other behavioral entities defined on the same scale. This is true even for Huntington’s disease. It is tempting to say that the choreiform movements of the patient with Huntington’s disease are caused by the HTT gene on chromosome 4, but doing so risks explaining the cause of an entity’s behavior with its own structure on a different scale. The turning of gears in a clock does not cause it to tell time: it is the clock telling time. Although it does not much matter in something as well-understood as Huntington’s disease, it is better to say that the cause of the disease is the inheritance of the HTT gene from an exogenous parent; the motor and cognitive symptoms that ensue are a larger-scale representation of the smaller-scale biological consequences of carrying the gene. The distinction between causal and compositional explanation is more important in entities that lack the well-understood structure of Huntington’s disease. Let’s say that an individual’s bankruptcy is caused by redlining practices in the local real estate market. The physical and behavioral entities that might make someone a target of redlining, such as socially defined race, economic class, and personal budget practices, all have blurry representations at smaller scales. So when redlining ⇒ bankruptcy, the whole blurry structure of the neurology of home ownership gets crossed with the equally blurry genetics of savings plans. That guarantees that the geneticist who conducts a genome-wide association study (GWAS) of bankruptcy will find something, but what is found will not be causal, and it will not explain anything. It will just be a blurry genetic representation of a process that can only be meaningfully analyzed at a larger scale. These examples illustrate why psychiatry’s old dichotomies are so persistent: they are not merely empirical puzzles waiting for better data, but reflections of the difficulty of drawing stable, meaningful boundaries around behavioral entities across multiple explanatory scales. Understanding the causal structure of complex behavioral entities requires a different skillset than the technical tools that are applied under the assumption that everything will eventually yield to biogenetic reduction. It requires philosophical discourse (this essay owes much to the philosopher of science W. Wimsatt4), advanced statistical methods (the clinical psychologist and philosopher P. Meehl devoted the second half of his career to the problem5), and (hardest of all), if recent history is any guide, attention to the null hypothesis. One must ask questions such as: “Are candidate gene studies of depression working, and what does it mean if they are not?“6. What it means is that the causal structure of depression may be more like the causal structure of bankruptcy than it is like Huntington’s disease. If that turns out to be the case, depression will fall more in the domain of clinical and social science as opposed to GWAS and brain imaging. The so-called biopsychosocial model is tautologically true in a hierarchical universe, but it obscures something important: entities are better defined, discussed and explained at some scales than they are at others. Determination of that optimal scale is prior to specification of causes. It must be understood what depression is before it can be understood what causes it. When we debate nature versus nurture, or mind versus brain, we are really debating how best to impose conceptual order on a system that resists tidy partitions. There may be no final resolution to these tensions, but there can be progress if we remain vigilant about the scales at which explanations operate, and maintain a distinction between genuine explanatory reduction, which is rare but powerful when it occurs, and simple compositional analysis, which is universally possible but generally uninformative.
The heritability (G) of cognitive ability increases substantially across development. However, previous studies of G x Age interaction—also known as the Wilson Effect—have not controlled for developmental changes in the reliability of cognitive assessment. Unmodeled reliability changes may bias estimates of cognitive growth and G x Age interaction, obscuring the developmental etiology of cognitive ability, but this possibility has not been explored empirically due to a lack of appropriate data and statistical methods. Using an expanded version of the Louisville Twin Study data set in which Ronald Wilson (1983) originally documented G x Age interaction, we first replicated heritability increases between ages 3 months and 15 years. Then, we examined the extent to which unmodeled reliability changes biased estimates of G x Age interaction using continuous time dynamic modeling. Our-best fitting model, which corrected for reliability differences across test battery and age, estimated that heritability increased from .26 to .55. Reliability increased substantially across development, from less than .30 in infancy to approximately .80 at age 15. Models that did not control for developmental reliability increases yielded downwardly biased estimates of cognitive growth and differentiation, G x Age interaction, developmental decreases in shared environmental effects, and increases in non-shared environmental effects. Results indicated that a re-evaluation of previous G x Age estimates may be warranted, and that future studies should control for developmental fluctuations in reliability. Our continuous time dynamic models improved estimation of developmental change in cognitive ability and can be easily adapted to other psychological constructs.
OBJECTIVES: Birth weight is a widely used indicator of prenatal experiences in models of the developmental origins of cognitive ability across the lifespan. This study aimed to examine the association between birth weight and cognitive ability using a community sample of twins followed prospectively from infancy to midlife. We leveraged the twin study design to identify phenotypic and biometric associations between the two constructs. METHODS: The sample consisted of 1,501 participants (387 dizygotic pairs, 360 monozygotic pairs, and 7 singletons; 53.1% female; 91.1% White) from the Louisville Twin Study. We modeled the change in the strength of the association between birth weight and cognitive ability using exponential decay functions. RESULTS: The magnitude of the association between birth weight and cognitive ability declined exponentially from infancy (β = .59, p < .05) to midlife (β = .27, p < .05). The lower asymptote of the exponential decay function was reached at about age 2.5 years of age, after which the association between birth weight and cognitive ability stabilized and remained constant up to midlife. A 1-kilogram increase in birth weight was associated with an 8.85-point increase in cognitive scores at 3 months and a 4.05-point advantage after about 2.5 years. Biometric regression models revealed that shared environmental factors accounted for the decline in the association between birth weight and cognitive ability. A small, positive within-pair association persisted into midlife. DISCUSSION: These findings suggest that prenatal experiences may have lasting effects on cognitive development across the lifespan, supporting developmental origin models of cognitive ability.
bioRxiv (Cold Spring Harbor Laboratory) · 2025-12-13 · 1 citations
articleOpen access
Previous brain-wide association studies (BWAS) cross-sectionally linked a specific behavioral trait, most commonly IQ or psychopathology, to variation in brain function or structure. Here, we expanded the focus of BWAS from effect sizes to interpretability and generalizability by mapping 649 variables to brain function and structure. We compared the resultant BWAS maps to other types of brain data to annotate the BWAS patterns. Socioeconomic status (SES) - not IQ or psychopathology - showed the strongest associations with both resting-state functional connectivity (RSFC) and cortical thickness in the Adolescent Brain Cognitive Development (ABCD) Study. A principal exposome brain pattern, anchored to sensory and motor cortex, captured 34% of the variance across all BWAS maps. This exposome pattern was strongly correlated with the SES and IQ BWAS maps and non-BWAS maps of sleep (EEG), norepinephrine (PET), and stimulants (drug trial), but not cognitive activation maps (task fMRI). Adjusting for SES, reduced brain-IQ associations by 40%. Brain with IQ associations did not generalize, as they could no longer be detected in subsamples drawn from only higher SES backgrounds, while brain with SES associations remained strong in higher-IQ-only subsamples. These findings reveal SES as the principal axis of population-level brain variation, possibly stemming from the sleep deprivation and heightened stress associated with lower SES, since socioeconomics can only indirectly affect the brain.
BACKGROUND: Clarifying risk factors for Alzheimer's disease (AD) in midlife permits intervening earlier in the lifespan and delaying conversion to AD. Understanding the relationship between blood-based AD biomarkers and memory functioning during middle age may help clarify whether elevated levels correspond to poor memory performance in later life. Blood-based biomarkers, including amyloid-β42/amyloid-β40 and ptau181, are associated with AD diagnosis, but studies investigating the correlates of these biomarker levels in middle age are scarce (Li et al., 2022; Karikari et al., 2020). METHOD: Edition (CVLT-3) were collected from 154 midlife twins (78 complete families) as part of the Louisville Twin Study. Pearson correlations and mixed effects regression analysis were used to estimate between-family and within-family associations between AD biomarkers and four CVLT-3 subtests (immediate word recall, short-delay free recall, long-delay free recall, and recognition). RESULT: Correlations of amyloid-β42/amyloid-β40 and the CVLT3 subtests ranged from -.02 to .04 while correlations with ptau181 ranged from -.09 to -.04. Although no correlations were statistically significant, the pattern of associations between episodic memory and ptau181 at least suggested a reliable negative association. Neither between-family nor within-family correlations were observed. CONCLUSION: Although AD biomarker levels have been found to correlate with memory functioning in older adult samples, we did not observe these same effects in middle adulthood. Thus, the current results suggest that individual differences in blood-based AD biomarkers may not have predictive utility for observed memory functioning in middle adulthood.
For decades, researchers have debated whether the magnitude of the Flynn Effect—intergenerational increases in mean IQ scores—varies across cognitive domains and subdomains, and whether the Flynn Effect reflects gains in general cognitive ability (g). We conducted a cross-domain, subtest-level investigation of the Flynn Effect across middle childhood and early adolescence (ages 7-15 years, N = 1187, 89% White, 9% Black, 52% female) in longitudinal cognitive ability data collected prospectively between 1957 and 1999 using three versions of the Wechsler Intelligence Scale for Children. Results provided clear evidence of the Flynn Effect as both increases in mean IQ score across generational cohort and decreases in mean IQ across test versions. Flynn Effect magnitude differed substantially across domains and subtests. Performance IQ gains were larger than full-scale and verbal IQ gains in cohort-based analyses, but test version-based estimates showed an opposite pattern (VIQ &gt; FSIQ &gt; PIQ). Variance in Flynn Effect magnitude across subtests did not follow a discernible pattern. The strength of the Flynn Effect on individual subtests was not strictly proportional to each subtest’s g-loading, with performance IQ subtests showing larger increases than would be expected given their g-loadings and verbal subtests showing smaller-than-expected gains.
DNA methylation age (DNAmAge) surpasses chronological age in its ability to predict age-related morbidities and mortality. This study analyzed data from 287 middle-aged twins in the Louisville Twin Study (mean age 51.9 years ± 7.03) to investigate the effect of DNAmAge acceleration on change in IQ (ΔIQ) between childhood and midlife, while testing childhood socioeconomic status (SES) as a moderator of the relationship. DNAmAge was estimated with five commonly used algorithms, or epigenetic clocks (Horvath, Horvath Skin and Blood, GrimAge, and PhenoAge). A factor analysis of these measures produced a two-factor structure which we identified as first generation and second generation measures. Results of genetically informed, quasi-causal regression models indicated that accelerated second generation DNAmAge predicted more negative ΔIQ from childhood to midlife, after accounting for genetic and environmental confounds shared by twins. The relationship between DNAmAge and ΔIQ was moderated by childhood SES, with a stronger effect observed among twins from low SES backgrounds. Second generation DNAmAge measures trained to estimate phenotypic biological age show promise in their predictive value for cognitive decline in midlife. Our genetically informed twin design suggested that epigenetic aging may represent a pathway through which early-life socioeconomic disadvantage impacts midlife cognitive health.
The conceptual and symptom overlap between depression and loneliness is so strong that depressive symptom rating scales often include measures of loneliness and their linear association ranges between .40 and .68. Despite these findings, few genetically informed studies have investigated the genetic and environmental association between them. In this brief report, we used 416 individual twins, 229 MZ twins and 187 DZ twins, from the Louisville Twin Study (MAge = 48.17; 56% female) to test two hypotheses: first, depressive symptomatology and loneliness are correlated genetically and environmentally; and second, the genetic correlation in women would be larger than men. Results from a correlated ACE factors model indicate that the genetic correlation is .52 and the nonshared environmental correlation is .27. No sex differences were observed. The strong genetic overlap between depressive symptomatology and loneliness raises questions about the biological mechanisms associated with common sets of genes.
