Understanding NULISAseq Data

Before diving into data analysis, it’s important to understand how NULISAseq technology works and how the data is generated and normalized.

What is NULISA™ Technology?

NULISA (NUcleic acid-Linked Immuno-Sandwich Assay) is a highly sensitive protein quantification platform that combines immunoassay specificity with next-generation sequencing readout.

The NULISAseq platform uses a proprietary dual selection proximity ligation approach to obtain high sensitivity and high signal-to-noise ratio for multiplexed protein detection:

Key Components:

  1. Verified antigen-specific antibody pairs - Two antibodies that bind different epitopes on the target protein
  2. Antibody-specific barcodes - Unique DNA sequences attached to each antibody
  3. Bridging oligos - Connect the barcodes only when both antibodies bind the same target
  4. Biotinylated oligos - Enable sequential capture on beads
  5. PolyA-tailed oligos - Allow for sequencing readout

Technical Workflow

The assay follows these steps:

This dual selection removes assay background and drastically improves signal-to-noise ratio.

Learn More About NULISA™ Technology

For a comprehensive understanding of the NULISA™ platform, we recommend:

NULISA™ Publication
Feng, W., Beer, J.C., Hao, Q. et al. (2023). “NULISA: a proteomic liquid biopsy platform with attomolar sensitivity and high multiplexing.” Nature Communications 14, 7238.
Read the full article

NULISA™ Platform Overview
Watch the NULISATM Platform Video for a visual explanation of the technology and workflow.

Technical Documentation
For detailed technical notes on assay design, normalization methods, and quality control:
See Technical Documentation in the Additional Resources chapter.

Available NULISAseq Panels

Alamar Biosciences offers several curated panels for disease-specific research, including:

  • Inflammation Panel: 250+ immune response markers including 120+ cytokines and chemokines
  • CNS Disease Panel: 120+ neuro-specific and neuro inflammation-related proteins
  • Mouse Panel: 120+ inflammation, neuro-degeneration and oncogenesis proteins specific to mouse model
  • Custom assays: Develop your own biomarker assays with the NULISAqpcr Custom Assay Development Kit

Explore Current Panels

For detailed panel specifications, target lists, and custom assay options, visit the Alamar Biosciences website

Data Normalization

NULISAseq uses a multi-step normalization process to make protein measurements comparable across samples and plates.

Raw Data: Sequencing Counts

The output from the sequencer is read counts for each target protein in each sample:

  • Each protein has a unique barcode
  • More reads = more protein present
  • Raw counts range from 0 to millions

Step 1: Internal Control (IC) Normalization

Purpose: Correct for sample-to-sample technical variation

Method:

  1. Each sample includes internal control spike-ins
  2. Divide each target’s read count by the sample’s IC count

Formula:

\[ \text{IC} - \text{normalized count} = \text{Raw count} / \text{IC count} \]

Step 2: Inter-Plate Control (IPC) Normalization

Purpose: Correct for plate-to-plate variation

Method:

  1. Calculate target-specific medians across 3 inter-plate controls (IPCs)
  2. Divide IC-normalized counts by these IPC target-specific medians
  3. Rescale by multiplying by 10⁴

Formula:

\[ \text{IPC} - \text{normalized count} = (\text{IC} - \text{normalized count} / \text{IPC median}) × 10⁴ \]

Step 3: Log2 Transformation

Purpose: Create NPQ (NULISA Protein Quantification) values in log2 scale

Method:

  1. Add 1 to all values (avoid log(0))
  2. Take log2 transformation

Formula:

\[ \text{NPQ} = log_2(\text{IPC} - \text{normalized count} + 1) \]

Special normalization for high-abundance targets

Most targets in NULISAseq panels are measured at their lower limit of detection. However, certain targets have exceptionally high endogenous levels that would saturate standard detection:

High-abundance targets:

  • Inflammation Panel: CRP and KNG1 in human plasma and serum samples
  • CNS Panel: APOE and CRP in human plasma and serum samples
  • Mouse Panel: Crp in mouse plasma and serum samples

Special tuning strategy: To accommodate both high-abundance and low-abundance targets in the same panel, these targets use a designed algorithm during normalization.This transformation is applied before the log₂ transformation step.

Key Note:

  • The listed LOD for these targets represents the upper limit of detection
  • NPQ values are valid and comparable for downstream analysis
  • Detectability is not a concern given the naturally high levels of these proteins
  • For novel sample types beyond validated matrices, the algorithm may not apply if abundance levels differ significantly, and NPQ may not be reported

Why Log Transform?

Using NPQ (log2-transformed values) instead of raw counts has many advantages:

✅ Stabilizes variance - Makes data more homoscedastic
✅ Reduces skewness - Data becomes more normally distributed
✅ Linearizes relationships - Easier to model
✅ Improves interpretability - Differences = fold changes
✅ Compresses range - Large values don’t dominate
✅ Reveals clearer patterns - Easier to see biological signals

NPQ vs Fold Change

Understanding the Relationship

Fold Change is calculated as:

\[ \text{Fold Change} = 2^{(\text{Difference in NPQ})} \]

Example:

Sample ΔNPQ Fold Change
1 2.3 5
2 4.1 17
3 5.7 53
4 7.3 161
5 8.9 485
6 10.5 1457

Key Point: NPQ differences are linear and easier to interpret than exponential fold changes!

Working with NPQ Values

While you cannot directly compare abundances across targets using NPQ, it is possible to calculate fold changes or ratios in relative abundances between individuals or groups.

Calculating Fold Changes from NPQ

If you need fold changes for interpretation:

# Given NPQ values
npq_disease <- 7.0
npq_healthy <- 4.0

# Calculate difference
delta_npq <- npq_disease - npq_healthy  # 3.0

# Convert to fold change
fold_change <- 2^delta_npq  # 2^3 = 8

Interpretation: Protein is 8 times higher in the disease group compared to healthy.

Comparing Across Targets

NPQ values can be compared across targets because:

  • All targets go through same normalization
  • Values are on same scale
  • Useful for calculating protein ratios

Example use case: Aβ42/Aβ40 ratio for Alzheimer’s diagnosis

Sample Target NPQ
Sample A (Healthy) Aβ42 9
Sample A (Healthy) Aβ40 6
Sample B (Alzheimer’s) Aβ42 5
Sample B (Alzheimer’s) Aβ40 7

Calculating Protein Ratios

Since NPQ values are log₂-transformed, calculating ratios is straightforward: subtract the NPQ values.

Formula for comparisons:

\[ log_2(A/B) = log_2(A) - log_2(B) \] \[ log_2(A/B) = \text{NPQ}_A - \text{NPQ}_B \] Calculations:

For Sample A (Healthy control):

  • NPQ(Aβ42) = 9, NPQ(Aβ40) = 6
  • Log₂(Aβ42/Aβ40) = 9 - 6 = 3
  • Actual ratio = 2³ = 8

For Sample B (Alzheimer’s patient):

  • NPQ(Aβ42) = 5, NPQ(Aβ40) = 7
  • Log₂(Aβ42/Aβ40) = 5 - 7 = -2
  • Actual ratio = 2⁻² = 0.25
Sample NPQ Aβ42 NPQ Aβ40 Log₂ Ratio Ratio
Sample A (Healthy) 9 6 3 8.00
Sample B (Alzheimer’s) 5 7 -2 0.25

Interpretation:

  • Healthy brain: Aβ42/Aβ40 ratio = 8.0
  • Alzheimer’s brain: Aβ42/Aβ40 ratio = 0.25

The Aβ42/Aβ40 ratio reflects the amyloid pathology, making this ratio a powerful biomarker for Alzheimer’s disease diagnosis and monitoring.

Data Analysis FAQ

1. How is the limit of detection (LOD) defined for NULISAseq?
Why do some targets have LOD of 0?

  • LOD is calculated separately for each analyte / target using the negative control (NC) samples.
  • For NULISAseq, the LOD is defined as follow:

Formula: \[ \text{LOD} = mean(\text{NC}) + 3 × sd(\text{NC}) \]

  • LOD is calculated on the normalized count scale, and then log2-transformed to be on the NPQ scale.
  • LOD = 0: Occurs when negative controls show no detectable signal at the current NGS depth.

2. Should I exclude samples below LOD?

We don’t recommend excluding individual values below LOD or replacing NPQ values with LOD.

Why not?

  • Excluding or replacing values creates skewed, non-normal distributions
  • Violates statistical assumptions for model assumptions for analysis
  • Reduces statistical power

Instead, we recommend using different detectability cutoff to exclude targets with low detectability

3. How do I calculate coefficient of variation (CV) for samples?

IMPORTANT: Unlog NPQ values before calculating CV.

NPQ values are in log₂ scale, but CV is only meaningful on the linear scale.

Step-by-step process:

  1. Back-transform NPQ to linear scale: \[ \text{Linear value} = 2^{\text{NPQ}} - 1 \]

  2. Calculate mean and standard deviation on linear scale: \[ \text{Mean}_{\text{linear}} = \text{mean(linear values)} \] \[ \text{SD}_{\text{linear}} = \text{sd(linear values)} \]

  3. Calculate CV: \[ \text{CV} = \frac{\text{SD}_{\text{linear}}}{\text{Mean}_{\text{linear}}} \times 100\% \]

Example in R:

# For replicate samples of a single target
npq_values <- c(8.5, 8.7, 8.3)  # NPQ values (log2 scale)

# Back-transform to linear scale
linear_values <- 2^npq_values - 1

# Calculate CV
mean_linear <- mean(linear_values)
sd_linear <- sd(linear_values)
cv_percent <- (sd_linear / mean_linear) * 100

print(paste0("CV = ", round(cv_percent, 2), "%"))
#> [1] "CV = 13.87%"

4. Can I compare data across different runs?

Yes! IPC normalization is designed to make data comparable across runs.

Quality metrics supporting cross-run comparability include:

  • Mean and median IPC CVs typically <10%
  • High inter-sample correlations (inter-run Pearson correlation ~ \(r = 0.95\))

In some cases intensity normalization or bridge samples (3-5 samples) can be used for further normalization. This enables longitudinal studies and meta-analyses.

5. How do you assess correlation between NPQ and absolute quantification data from other platforms?

Transform datasets from other platforms to log2 scale first!

Recommended approach:

  • Convert absolute quantification values (e.g., pg/mL) to log2 scale
  • Compare log2-transformed values against NPQ values
  • Visualize with scatterplots before choosing correlation method

Choosing the appropriate correlation method:

✅ Use Pearson correlation (r) when:

  • Scatterplot shows approximately linear relationship
  • No extreme outliers present
  • Data appears normally distributed

✅ Use non-parametric methods (Spearman ρ or Kendall τ) when:

  • Scatterplot shows nonlinear trend
  • Extreme outliers present
  • Very small sample size
  • Data far from normal distribution

Note: For non-parametric methods, the specific data transformation is less critical, but log2 transformation still helps with visualization and interpretability.

Continue to: Chapter 1: Data Import