Supporting data for "Polyethersulfone-based adsorptive membranes for dissolved pollutants: closing the gap between single-solute performance and complex-matrix operation"
Description
Structured performance database underlying the review article "Polyethersulfone-based adsorptive membranes for dissolved pollutants: closing the gap between single-solute performance and complex-matrix operation". CONTENT. One row per membrane variant and target pollutant with quantitative data: 559 rows in 34 fields, drawn from 121 publications. 499 rows were extracted from the full text of 61 studies; 60 rows were taken from abstracts and are flagged in the data_source field. Fields cover the scaffold polymer, the active phase and its loading, the fabrication route and membrane format, the target pollutant and feed type, the matrix conditions tested, isotherm capacity with its normalization basis, removal or rejection, flux and permeance, fouling ratios, regeneration cycles and retained performance, leaching, and the reported capture mechanism. A second table lists the 35 clean-water versus complex-matrix comparisons behind the retention ratio rho used in the article, one row per pair. FILES. PES_adsorptive_membrane_database_v2.0.xlsx (README sheet, the 559-row table, the retention-ratio pairs, the field list); PES_adsorptive_membrane_database_v2.0.csv (the 559-row table alone, UTF-8); rho_pairs_v2.0.csv (the 35 pairs); README.md. READ BEFORE USE. (1) Filter on data_source: abstract-only rows are excluded from every statistic reported in the article. (2) Do not pool qmax_mg_g across values of qmax_basis; 65 of the 120 full-text capacity values carry no stated normalization basis, and their median is about 2.5 times that of the values normalized to the whole membrane. (3) Rows with loading_wt_pct = 0 are pristine-polymer controls, not adsorptive membranes. (4) Three unit normalizations were applied and no others: flux to permeance by division by the stated transmembrane pressure; MPa and kPa to bar; g/L and ppb to mg/L. VERIFICATION. Every value used in the article was checked against its source publication. In addition, a seeded random sample of 50 of the 499 full-text rows (24 studies) was re-extracted independently against the sources: of 237 numbers in ten quantitative fields, 234 matched exactly and three differed only by the unit normalizations above. The field-level transcription-error rate in the sample is 0 of 237, with a 95% upper bound of 1.3% by the rule of three. Free-text fields and abstract-only rows were not audited. KNOWN BIASES. Publication bias against membranes that failed in a complex matrix; a literature search restricted to CrossRef, Semantic Scholar and Europe PMC; single-screener selection; and some capacity values reproduced from the summary tables of reviews.
Files
Steps to reproduce
1. Search. CrossRef, Semantic Scholar and Europe PMC were queried through their public application programming interfaces in June 2026 and again in September 2026, using thirty free-text relevance queries against the combined title, abstract and keyword metadata, capped at sixty records per query per source. The thirty queries are listed verbatim in Supplementary Note S1.1 of the article. Retrieval returned 1,375 unique records after de-duplication by DOI. The request was limited to 2010-2026; screening was restricted to English-language peer-reviewed journal articles, with no grey literature. 2. Screening. Records were screened on title and abstract against the inclusion and exclusion criteria of Supplementary Note S1.2, leaving 186 publications in the corpus. Full text was sought and obtained for 77 primary studies, of which 61 yielded quantitative rows. Screening and full-text assessment were carried out by the first author. 3. Extraction. Each membrane variant and each target pollutant with quantitative data was entered as one row in the 34 fields listed on the "fields" sheet. Values were transcribed as reported. Only three unit normalizations were applied: flux to permeance by division by the transmembrane pressure stated in the same row (rows reporting flux at an unstated pressure were left without a permeance value); MPa and kPa to bar; and g/L and ppb to mg/L. A further 60 rows were taken from the abstracts of 60 publications for which full text could not be obtained; those rows carry "abstract only" in the data_source field and are excluded from every statistic reported in the article. 4. Retention ratios. Every clean-water versus complex-matrix comparison in the corpus was extracted into the "rho_pairs" sheet, one row per pair, with the endpoint type and matrix class recorded. A pair enters the pooled retention ratio only when both values are numerical and the two differ in feed composition alone; comparisons that differ in feed concentration or that follow regeneration are listed but flagged and pooled separately. 5. Verification. Every value used in the article was checked against its source publication. In addition, 50 of the 499 full-text rows were drawn with a fixed random seed (20260911, without replacement, in database order) and re-extracted independently against the sources; the sample spans 24 studies. Of 237 numbers in ten quantitative fields, 234 matched exactly and three differed only by the unit normalizations in step 3. The search and screening can be re-run from the queries and criteria above. The extraction can be checked row by row against the DOI recorded in each row.