Predictors of Reading (Dis)Ability

Purpose

Reading ability is a complex neurocognitive skill that develops through interactions between genetic, environmental, and behavioral factors. Dyslexia, a common reading disorder, affects approximately 5–15% of children and has substantial heritability, yet its underlying mechanisms remain incompletely understood.

Traditional research approaches have several limitations:

Reading ability is often reduced to single measures (e.g., word reading)
Genetic studies rely on one-variant-at-a-time methods (GWAS)
Studies rarely integrate genetic and non-genetic predictors

The goal of this project was to develop and apply a comprehensive analytic pipeline to identify predictors of reading (dis)ability by integrating:

Multidimensional reading phenotypes
Genome-wide genetic data
Demographic, environmental, and behavioral factors
Machine learning methods

What is reading (dis)ability?

Reading is not a single skill—it reflects the integration of multiple component processes, including:

Phonological processing
Word decoding
Language comprehension
Attention and cognitive control

Children with reading disorders may show different profiles:

Dyslexia (decoding difficulties)
Poor comprehension (language-related deficits)
Comorbid profiles (both)

To better capture this complexity, we modeled reading ability as a latent trait, combining multiple reading measures into a single, theory-driven construct.

What do we already know about predictors?

Reading (dis)ability arises from interacting systems, not isolated factors.

Genetic influences

Reading is polygenic, influenced by many small genetic effects
Known genes (e.g., KIAA0319, DCDC2) interact within biological pathways
Traditional GWAS often identifies few significant SNPs due to strict corrections

Behavioral and demographic influences

Language skills (vocabulary, receptive language)
Nonverbal IQ
Maternal education
Birth outcomes and environmental exposures

Key gap

Most studies isolate these domains instead of modeling their combined effects—a limitation this project directly addresses.

How we did it

Data source: ALSPAC

We used the Avon Longitudinal Study of Parents and Children (ALSPAC), a population-based birth cohort that includes genetic, behavioral, and environmental measures across development.

We used:

Behavioral data from ages 7, 8, and 9
Genome-wide genetic data for thousands of participants
Survey data from birth and early life

Final analytic sample: ~3,232 children with complete genetic + behavioral data

Integrated analytic pipeline

Note. Pipeline to identify predictors of reading ability.

We developed a four-step analytic pipeline:

1. Latent phenotype construction

Used confirmatory factor analysis (CFA)
Combined multiple measures (word reading, spelling, comprehension)
Reduced measurement error and improved statistical power

2. SNP screening

Combined:
- Prior candidate gene findings
- Genome-wide association results
Identified a set of informative SNPs for modeling

GWAS identified:

1 genome-wide significant SNP (rs181384543) in PTCHD1-AS

3. Machine learning (Elastic Net)

We used elastic net regression to model many predictors simultaneously:

Handles high-dimensional genomic data
Selects informative features
Captures correlated and interacting predictors

Two models:

SNP-only model
SNP + demographic/behavioral model

4. Pathway enrichment analysis

We mapped identified SNPs to genes and biological pathways:

Identified overrepresented biological processes
Linked genetic findings to brain development mechanisms

What we learned

1. Reading ability is best modeled as a system

Using a latent phenotype:

Improved detection of genetic signals
Reduced measurement error
Produced more stable and interpretable results

2. Genetic effects are distributed and additive

GWAS alone identified 1 significant SNP
Elastic net identified:
- ~62 SNPs (SNP-only model)
- ~96 SNPs (combined model)

➡️ Insight: Reading ability reflects many small genetic effects, not monogenic origins (Lancaster et al., 2020).

3. Non-genetic factors dramatically improve prediction

SNP-only model explained ~12% of variance
Full model explained ~32% of variance

Important positive predictors included:

Vocabulary (r ≈ .42)
Nonverbal IQ (r ≈ .31)
Receptive language (r ≈ .29)
Mother’s education (r ≈ .25)

➡️ Insight: Behavioral and environmental factors are essential, not optional (Lancaster et al., 2020).

4. Biological pathways—not single genes—drive reading

We found enrichment in pathways related to:

Neuron and dendrite development
Synaptic and postsynaptic structures
Learning and memory processes

➡️ Reading ability emerges from interconnected neural systems (Lancaster et al., 2020).

5. Cognitive architecture differs by reading ability

From complementary analyses:

Typical readers:
- Attention influences reading indirectly
Children with reading disorders:
- Attention directly influences reading comprehension

➡️ Insight: The structure of reading changes with impairment, not just performance level (Lancaster et al., 2021).

Why this work matters

This project introduces a new framework for studying complex developmental traits:

✅ Moves beyond single-gene models
✅ Integrates genetic + behavioral + environmental data
✅ Uses machine learning to identify meaningful patterns
✅ Improves phenotype definition using latent constructs
✅ Links genetic findings to biological pathways

Broader implications

Better early identification of children at risk
More precise, individualized intervention strategies
Improved theoretical models of reading development
Generalizable pipeline for other complex traits

Team & support

This work was supported by NIH F32 grant (1F32HD089674-01A1) to Dr. Hope Lancaster.

Dr. Hope S. Lancaster

Conceptualization, analyses, modeling, writing, funding

Mentorship team:

Dr. Jing Li — machine learning & AI
Dr. Shelley Gray — reading and language
Dr. Valentin Dinu — biostatistics

Posters & Downloads

Comparison of two phenotypes to identify SNPs associated with poor reading
Lancaster, Dinu, & Li - ASHG 2018
Download the poster (PDF)

Predicting reading comprehension in children with different types of reading disorders
Lancaster, Li, & Gray - SSSR 2018
Download the poster (PDF)

Sparse machine learning model and pathway analysis suggest novel genetic associations with dyslexia
Lancaster, Liu, Dinu, & Li - BGA 2019
Download the poster (PDF)

The purpose of this study was to examine the relationship between selective visual attention (SVA), reading decoding, listening comprehension and reading comprehension in children with and without a reading disorder. We used longitudinal data from the Avon Longitudinal Study of Parents and Children. We split children into four groups: Typical Readers, Dyslexics, Poor Comprehenders and Comorbid Reading Disorder. We included measures of single word reading, nonword reading, spelling, phonological processing, vocabulary, receptive language, nonverbal intelligence, selective attention and reading comprehension. We used analysis of variance, correlations and structural equation modelling to examine the relationship between SVA and reading. We fit two possible models: SVA Indirect and SVA Direct. The difference between these models was the inclusion of a direct path from SVA to reading comprehension. We examined an indirect model, where SVA predicted reading comprehension through word decoding and listening comprehension, and a direct model, which included a pathway from SVA to reading comprehension. Based on our analysis of variance and correlation results, we collapsed the Dyslexic, Poor Comprehenders and Comorbid Reading Disorder groups for the structural equation modelling. We found evidence that for Typical Readers, an indirect model was the best fit, whereas the direct model was the best model for children with a reading disorder. Selective visual attention is related to reading comprehension. This relationship differs for children with and without a reading disorder.

Past research has suggested that reading disability is a complex disorder involving genetic and environment contributions, as well as gene–gene and gene–environment interaction, but to date little is known about the underlying mechanisms. Using the Avon Longitudinal Study of Parents and Children, we assessed the contributions of genetic, demographic, and environmental variables on case–control status using machine learning. We investigated the functional interactions between genes using pathway and network analysis. Our results support a systems approach to studying the etiology of reading disability with many genes (e.g., RAPGEF2, KIAA0319, DLC1) and biological pathways (e.g., neuron migration, positive regulation of dendrite regulation, nervous system development) interacting with each other. We found that single nucleotide variants within genes often had opposite effects and that enriched biological pathways were mediated by neuron migration. We also identified behavioral (i.e., receptive language, nonverbal intelligence, and vocabulary), demographic (i.e., mother’s highest education), and environmental (i.e., birthweight) factors that influenced case–control status when accounting for genetic information. The behavioral and demographic factors were suggested to be protective against reading disability status, while birthweight conveyed risk. We provided supporting evidence that reading disability has a complex biological and environmental etiology and that there may be a shared genetic and neurobiological architecture for reading (dis)ability.