Flow diagram of exploring RPLS dichotomous classification

Baseline characteristics of training cohort and validation cohort

Cell cycle, DNA damage and repair, and metabolism are dysregulated in RPLS

Volcano plot of the DEGs in 8 normal vs 8 RPLS tissues (A). Venn diagram showed shared genes between DEGs and prognostic genes (B). GSEA analysis of RPLS tumors, including HALLMARK gene sets and REACTOME gene sets (C). Circular plots of the prognostic genes in GO, KEGG, and enrichWP (D).

RPLS subgroups (G1 and G2) based on cell cycle, DNA damage and repair, and metabolism.

tSNE exhibited the subgroups (G1 and G2) of RPLS (A). Survival cures of OS (B) and DFS (C) in G1 and G2. The hierarchical clustering heatmap of dysregulated pathways in G1 and G2 (D). Histograms revealed the difference of pathological composition ratio (E), surgery times (F), and MDM2 (G) in G1 and G2. Violin plot of the microenvironmental scores in G1 and G2 (H).

RPLS classification strategy (C1 and C2) derived from RPLS subgroups.

NMF for a re-classification of training cohort 1 (C1 and C2) (A). Survival cures of OS (B) and DFS (C) in C1 and C2. Histograms revealed the difference in pathological composition ratio (D), MDM2 (E), and surgery times (F) in C1 and C2.

RPLS dichotomous classification (Cluster_C1 and Cluster_C2) derived from RPLS clusters.

Heatmap of biomarkers identified (LEP and PTTG1) in C1 and C2 (A). ROC curves of the machine learning models to identify C1 and C2 (B). Correlation between LEP and PTTG1 expression (C). Survival curves of OS (D) and DFS (E) in Cluster_C1 (low risk) and Cluster_C2 (high risk) groups. GSEA of HALLMARK gene sets in Cluster_C1 and Cluster_C2 (F). Histograms revealed the difference of pathological composition ratio (G), MDM2 level (H), and surgery times (I) in Cluster_C1 and Cluster_C2. Sankey diagram indicated the correlation among G1/G2, C1/C2, and Cluster_C1/C2 (J).

Validation of the RPLS dichotomous classification in another 241 RPLS cohort

Representative IHC staining images of LEP (A) and PTTG1 (B). Survival curves of OS (C) and DFS (D) in high-risk and low-risk groups.

Survival nomogram of LEP+PTTG1 model in validation cohort.

The difference of surgery times (A), pathological composition (B), and surgery times (C) in high-risk and low-risk groups. Nomograms for OS was developed in REASR cohort with four factors: sex, age, risk score, and differentiation (D). ROC curves of 1-, 2-, and 3-year OS in validation cohort (E). Calibration curves of predicting 1-, 2-, and 3-year OS in validation cohort (F).

Clinical features and re-classification of RPLS subgroups (G1 and G2).

The difference of Ki67 (A) and tumor size (B) in G1 and G2. Volcano plot of the DEGs (G1 vs G2) (C). Venn diagram showed shared genes between DEGs and prognostic genes (D). Bubble plot of the DEGs enrichment (E). NMF for a re-classification of training cohort 1 (F). tSNE exhibits the RPLS clusters (C1 and C2) (G).

Dysregulated pathways and clinical features of RPLS clusters and high-/low- risk groups.

The hierarchical clustering heatmap of dysregulated pathways in C1 and C2 (A). The difference of Ki67 (B) and tumor size (C) in C1 and C2. The hierarchical clustering heatmap of dysregulated pathways in high- and low-risk groups (D). The difference of tumor size (E) and Ki67 (F) in high- and low-risk groups.

Validation of LEP+PTTG1 model in an external liposarcoma cohort.

Survival curve of DRFS in low-risk and high-risk groups (A). ROC curves of 1-, 3-, and 5-year DRFS in validation cohort (B). Histograms revealed the difference of pathological composition ratio in low-risk and high-risk groups (C). Pathological type details of LPS patietns (D).

Survival nomogram of LEP+PTTG1 model in validation cohort.

Nomograms for DFS was developed in REASR cohort with four factors: sex, age, risk score, and differentiation (A). ROC curves of 1-, 2-, and 3-year DFS in validation cohort (B). Calibration curves of predicting 1-, 2-, and 3-year DFS (C).