Deciphering Parkinson's Disease:

From Biomarker Discovery to Therapeutic Insights

Proposal Defense: Nicholas Minster, M.S.

Committee Chair: Prof. M. Saleet Jafri
Committee Members: Prof. Ancha Baranova, Prof. James C. Thompson

Agenda

  • Background and Literature Review
  • Specific Aims and Significance
  • Research Approach
  • Methods
  • Data
  • Preliminary Results
  • Timeline
  • Conclusion
  • References

Background

Approximately 10 million global cases, doubled from 1990 to 2019.


Parkinson's Disease (PD) involves motor symptoms like tremor and rigidity, non-motor complications like cognitive decline, and progressive disability, severely impacting quality of life and increasing caregiver burdens.


Pathology: Neurodegeneration, Lewy bodies, and α-synuclein aggregation.


Diagnosis is based on clinical evaluation.

The distribution of Parkinson's disease incidence in age groups of 204 Countries from 1990 to 2019. (A) was the incident number in age groups. [1]

Literature Review



Biomarkers enable early, precise detection of Parkinson’s Disease.


Genetic mutations in GBA1 and LRRK2 are key risk factors for PD.


Advances in biomarker assays include α-synuclein amplification and fluid-based markers.


Multimodal data integration, combining genetic, proteomic, and imaging datasets, enhances understanding of PD heterogeneity, improves diagnostic precision, and supports personalized therapeutic strategies.

Syn-one biopsy immunofluorescence image taken at UCSD.

Gaps in the Field

  • Absence of disease-modifying therapies

    The current standard is supplemental dopamine.

  • Lack of validated prodromal biomarkers

    Current biomarkers rely on disease onset and targeting has not yet proven successful in preventing disease progression.

  • Absence of meaningful disease subtypes

    Stratification is typically based on qualitative grading, i.e. UPSIT scores.
Nalls

Nature 2022 Landmark Study [2]



This proposal builds on, improves, and expands on the work done in this seminal study. Preliminary work has already touched on the following points.

Shifted data modalities to include proteomics and exclude qualitative features like UPSIT scores, resulting in a more actionable feature set.


Improved machine learning methods and results.


Directly linked identified markers to drug datasets.


Included robust model validation practices to ensure model generalizability and future adaptability.

Specific Aims

These aims align with the objectives of major Parkinson’s research institutes and addresses current gaps in the literature

  • Aim 1:

    1. Identification and validation of multimodal biomarkers for PD.

  • Aim 2:

    1. Analysis of biomarkers' roles in biological pathways and disease progression.

  • Aim 3:

    1. Biomarker dynamics and perturbational analysis

  • Aim 4:

    1. Stratification and repeated analysis.
AMP-PD

Significance

Why This Research Matters

Bridge the gap between predictive modeling and biological insights.

Improve prodromal diagnosis and therapeutic interventions.

Address heterogeneity in early disease stages.

Research Approach

  • Theoretical Framework

    Alignment with goals of major research initiatives (e.g., PPMI, PDBP, BioFIND).

  • Data Sources

    Parkinson's Progression Markers Initiative (PPMI) and the Parkinson's Disease Biomarkers Program (PDBP).

  • Modalities

    RNA-seq, Proteomics, MRI images, and Clinical Data.

Methodology

Minster
Minster

Data

Statistics on PDBP and PPMI

The number of samples for each cohort-
PPMI: 960
PDBP: 657

After preprocessing the data set includes:
Ensemble gene IDs: 1238
Uniprot IDs: 1463

Additional data like Axial T2-weighted MRI: 45 patients

PDBP was selected as the training set for several reasons:

Includes participants at a wider range of stages.

Contains fewer samples but PDBP is higher quality and has more samples per patient.

Preliminary Results

Binary Modeling

The best performing model was the combined model.

Applied a pipeline comprising variance thresholding, feature selection, scaling, and modeling.

Used GroupKFold cross-validation with randomized hyperparameter optimization.

Selected 41 features: 12 Ensemble gene IDs and 29 Uniprot IDs using SelectKBest with ANOVA F-statistic.
Validation ROC/AUC Validation F1 Score
Transcriptomics 0.60 0.50
Proteomics 0.85 0.77
Combined 0.87 0.78

Analysis of the Combined Model Minster Minster

Analysis of the Combined Model Minster Minster

What Drugs Perturb this Signature

Tau scores +/- 90 are considered significant.


Positive tau pushes the targets in the disease direction. Negative towards the healthy controls.


Digitalis plant derivatives (foxglove)- Mediate development of PD through inhibiting ATPase, altering the transport of cations, metabolites, and neurotransmitters across cellular membranes [3].


SAL-1 (-95.06 tau)- A2A antagonists confer protection against the degeneration of dopaminergic neurons. [4]

Limitations and Challenges

  • Data

    Data collection strategies were not designed for machine learning.

    There is limited demographic representation.

    The target variable is determined subjectively.

  • Modeling

    Features with non-linear relationships may not be well-represented due to the selectkbest approach.


    Stratification assumes adequate sample sizes within subgroups, power calculations indicate feasibility.

Stratification

PCA

MRI Data Exploration





The correlation between LSMEM1 and MRI entropy suggests that changes in LSMEM1 (ENSG00000181016) expression or function may influence tissue microstructure and complexity.



These activities could alter cellular organization, heterogeneity, or integrity in tissues, leading to changes detectable as variations in MRI entropy.

Timeline

Timeline

Future Directions



Validation with other datasets.

Exploration of fusion models for patients with missing modalities.

Long-Term Goals:

Support clinical translation of findings.

Conclusion

This work represents a critical step towards advancing our understanding of Parkinson's Disease through integrative multi-omics analysis.

  • Expected Contributions:
  • Identification of novel biomarkers with potential diagnostic and therapeutic implications.
  • Enhanced stratification of disease subtypes to address patient heterogeneity.
  • Demonstrated the power of combined RNA-seq and proteomics data for predictive modeling.

Thank you for your attention. I welcome your questions and feedback.

Refrences

  • 1. Ou Z, Pan J, Tang S, Duan D, Yu D, Nong H and Wang Z (2021) Global Trends in the Incidence, Prevalence, and Years Lived With Disability of Parkinson's Disease in 204 Countries/Territories From 1990 to 2019. Front. Public Health 9:776847. doi: 10.3389/fpubh.2021.776847
  • 2. Makarious, M.B., Nalls, M. et al. Multi-modality machine learning predicting Parkinson’s disease. npj Parkinsons Dis. 8, 35 (2022). https://doi.org/10.1038/s41531-022-00288-w
  • 3. Kurup, R.K., Kurup, P.A., 2003. Hypothalamic digoxin-mediated model for Parkinson’s disease. Int J Neurosci 113, 515–536. https://doi.org/10.1080/00207450390162263
  • 4. Cieślak, M., Komoszyński, M., Wojtczak, A., 2008. Adenosine A2A receptors in Parkinson’s disease treatment. Purinergic Signal 4, 305–312. https://doi.org/10.1007/s11302-008-9100-8