I have RNAseq data (expected counts) from a large set of patients from GTEx and TCGA. However, I am only interested on analyzing the differential expression of a small set of genes (about 25). Can I download the expected counts values of only these genes and run DE analysis of only this set (which is more practical for me), of do I have to download the whole dataset and check if those 25 are differentially expressed?
No, as most of the statistics involved in DEG analysis (normalization, dispersion estimation...) assumes a large number of genes that are not changing between condition. Perform standard analysis as described in the manuals of e.g. DESeq2, edgeR, limma and then check those you are interested in. See if TCGA already provides a count matrix with raw counts to make your life easier. You need raw counts for a meaningful analysis, not TPM, FPKM etc. Please use the search function on starting DEG analysis from pre-normalized counts to learn why this is not recommended.