Deconvolution of Peripheral Blood Transcriptomes to Evaluate Immune Profiles Associated with Minimal

Residual Disease in Pediatric Acute Lymphoblastic Leukemia:

A Machine Learning-Driven Study

Husna Irfan Thalib1, Sariya Khan1, Ayesha Jamal1, Ammarah Tayyebah1

, Fatima Aslam2

1Batterjee Medical College, Jeddah, Saudi Arabia; 2Tucson Medical Center, Tucson, Arizona, USA

Poster Code: soho2026.64308cd

Background

Results

Conclusion

• THE CLINICAL PROBLEM: MRD is a

major prognostic marker in pediatric ALL

and guides treatment stratification.

• THE GAP: Advanced immune profiling

can be resource-intensive and may

have limited accessibility.

• THE APPROACH: Peripheral blood

transcriptomic deconvolution may

provide a scalable method to

characterize immune-cell profiles

associated with MRD.

• HYPOTHESIS: Immune-cell profiles

inferred from peripheral blood

transcriptomes differ by MRD status and

can discriminate MRD-positive from

MRD-negative patients.

• OBJECTIVE: To determine whether

machine learning–based immune

deconvolution of peripheral blood

transcriptomes can identify immune-cell

patterns associated with MRD status in

pediatric ALL.

Figure 1. Statistical significance (−log₁₀ p-value) of immune-

cell population differences between MRD-positive and MRD-

negative patients. Arrows indicate direction of change in the

MRD-positive group; dashed line marks α = 0.05

Figure 2. Receiver operating characteristic (ROC) curve for

the Random Forest classifier (AUC = 0.91, 95% CI 0.84–0.97).

• Peripheral blood transcriptomic

deconvolution identified distinct immune

profiles associated with MRD status in

pediatric ALL.

• MRD-positive patients demonstrated

increased regulatory T-cell and

monocyte proportions alongside

reduced CD8+ T-cell representation,

while a Random Forest classifier

achieved strong discrimination of MRD

status (AUC 0.91).

• These findings support further

investigation of transcriptomic immune

profiling as a potential adjunct to MRD

assessment, particularly in settings

where advanced immunophenotyping is

less accessible.

• All in all, they highlight the potential of

transcriptomic immune deconvolution to

uncover biologically meaningful

immune signatures associated with

residual disease in pediatric ALL.

Methodology

Table 1. Summary of immune-cell populations differentiating

MRD-positive from MRD-negative patients.

Clinical Implications

• Study design: Retrospective

computational analysis of publicly

available pediatric ALL transcriptomic

data.

• Cohort: GSE48558 | n = 78 pediatric

ALL patients. Peripheral blood gene-

expression profiles with MRD status

annotated in the original dataset.

Table 2. Random Forest classifier performance

metrics using CIBERSORTx-inferred immune-cell

proportions as features.

• Adjunctive: May complement

established MRD assessment

• Accessible: Uses peripheral blood

transcriptomic data

• Scalable: Potential for computational

application across datasets

• Biologically informative: Captures

MRD-associated immune patterns

Contact

Figure 4. T