Methods Section Guide

## Materials and Methods

### Cell Culture and Treatment
HeLa cells (ATCC CCL-2) were maintained in DMEM (Gibco, #11965092)
supplemented with 10% FBS (Gibco, #26140079) and 1% penicillin-
streptomycin (Gibco, #15140122) at 37C in 5% CO2. Cells were
seeded at 5 x 10^4 cells/well in 24-well plates and treated with
compound X (0.1, 1, 10 uM) for 24 hours.

### Western Blot Analysis
Total protein was extracted using RIPA buffer (Thermo, #89900)
with protease inhibitor cocktail (Roche, #04693116001). Proteins
(30 ug/lane) were separated on 10% SDS-PAGE gels and transferred
to PVDF membranes. Primary antibodies: anti-TargetProtein
(Cell Signaling, #1234, 1:1000), anti-beta-actin (Sigma, #A5441,
1:5000). Secondary antibodies: HRP-conjugated (1:10000).

## Methods

### Dataset
We evaluated our method on three benchmark datasets:
- **ImageNet-1K** (Russakovsky et al., 2015): 1.28M training images,
  50K validation images across 1,000 classes
- **CIFAR-100** (Krizhevsky, 2009): 50K training, 10K test, 100 classes
- **Oxford Flowers-102** (Nilsback & Zisserman, 2008): 8,189 images, 102 classes

### Model Architecture
Our model extends the Vision Transformer (ViT-B/16) with the
following modifications:
1. Replaced standard self-attention with linear attention (Katharopoulos et al., 2020)
2. Added a learnable class-conditional normalization layer after each block
3. Used patch size 16x16 with input resolution 224x224

### Training Details
| Hyperparameter | Value |
|---------------|-------|
| Optimizer | AdamW (beta1=0.9, beta2=0.999) |
| Learning rate | 1e-3 with cosine decay |
| Weight decay | 0.05 |
| Batch size | 256 (across 4 A100 GPUs) |
| Training epochs | 300 |
| Warmup epochs | 10 |
| Data augmentation | RandAugment (N=2, M=9), Mixup (alpha=0.8) |
| Label smoothing | 0.1 |

All experiments were implemented in PyTorch 2.1 and run on 4x NVIDIA A100
80GB GPUs. Training took approximately 18 hours per run. Code is available
at [repository URL].

## Methods

### Participants
A total of 412 participants (245 female, 162 male, 5 non-binary;
M_age = 34.2, SD = 11.8) were recruited via Prolific. Inclusion
criteria: (a) aged 18-65, (b) fluent in English, (c) resided in
the US. Exclusion criteria: (a) failed two or more attention checks,
(b) completed the survey in under 3 minutes. After exclusions,
387 participants remained (attrition: 6.1%).

Sample size was determined a priori using G*Power 3.1 (Faul et al., 2007).
For a medium effect size (f^2 = 0.15), alpha = .05, and power = .80
in a multiple regression with 5 predictors, the required sample was 92.
We oversampled to ensure adequate power for subgroup analyses.

### Measures

**Perceived Stress Scale (PSS-10)** (Cohen et al., 1983): 10 items,
5-point Likert scale (0 = never, 4 = very often). Cronbach's alpha
in the current sample: .87.

**Big Five Inventory (BFI-10)** (Rammstedt & John, 2007): 10 items,
5-point Likert scale. Subscale alphas ranged from .68 to .81.

### Procedure
After providing informed consent, participants completed measures in
the following fixed order: demographics, PSS-10, BFI-10, experimental
task, manipulation check, debriefing. Median completion time: 14 minutes.
Participants were compensated GBP 2.50.

### Ethical Approval
This study was approved by the [University] IRB (Protocol #2024-0123).
All participants provided informed consent.