github durrantlab/dimorphite_dl v2.1.0

one month ago

Dimorphite-DL 2.1.0

Changed

  • Truncation. When max_variants truncates the output, the most probable states are kept, scored by the Henderson-Hasselbalch fraction at the middle of the pH range.

    • Previously the first states generated were kept, and enumeration stopped there, so every later site stayed in its input state.
  • Logging. Console logging from enable_logging and --log_level now goes to stderr, so stdout carries only SMILES.

    • The stdout_set argument keeps its name.
  • Output order. Output order no longer depends on PYTHONHASHSEED; variants come out in enumeration order.

  • New narrow substructure entries. site_substructures.smarts now carries narrow entries, each placed ahead of the broader entry it is drawn from, for:

    • simple aliphatic amines

    • unactivated phenols

    • alkanethiols

    • primary sulfonamides

    • C-substituted amidines

    • N-alkyl anilines

    • simple carboxamides

    • tetrazoles

    • the nucleobases

      Effects of the new entries:

    • Compounds whose pKa lies well outside the pH range now resolve to a single state rather than both: phenol, ethanethiol, benzenesulfonamide, methanesulfonamide, methylamine, benzylamine, ethanolamine, piperidine, triethylamine, quinuclidine, propranolol, 2,2,2-trifluoroethylamine, acetamidine, benzamidine, N-methylaniline, benzamide, acetanilide, peptide backbone N-H, and 5-methyltetrazole, among others.

    • Amides and anilides are no longer deprotonated at physiological pH, so with max_variants=1 a molecule bearing both an amide and a genuinely acidic group (e.g., a tetrazole) is ionized at the acidic group.

    • This changes the output for existing inputs, so pipelines that depend on the previous enumeration should be checked.

    • The means and standard deviations fitted from the training data are unchanged; each added entry takes its values from published measurements, cited in the substructure file.

  • Benchmark. Measured against the external test sets of Baltruschat M, Czodrowski P (2020) F1000Research 9:113, scoring the annotated ionization site of 399 of their 403 compounds, the share returning only the dominant species at pH 7.4 rose from 48% to 65%.

    • Fifteen compounds omit the dominant species in both versions.
    • One more does so in 2.1.0: a fused pyrazole N-H with a measured pKa of 7.35, which the new azole entry places well above the pH range, although its anion is only about half the population at pH 7.4.
    • One compound, whose basic site no pattern covers, is never ionized in either version.
  • CLI identifiers. The CLI now appends each input's identifier to every variant it produces, as SMILES,identifier.

    • Names in the input file were dropped, so variants could not be matched back to the molecule they came from.
    • Scripts that parse CLI output should expect the comma-separated field; from Python, pass label_identifiers=True to protonate_smiles for the same output.
  • RDKit. RDKit 2026 releases are now supported; the dependency bound is raised from <2026 to <2027.

  • Identifiers with whitespace. An identifier may now contain whitespace.

    • Only the second whitespace-separated field was kept, so methyl phosphate was truncated to methyl, and a list item with more than two fields was rejected rather than read.

Fixed

Output and labeling

  • label_states and --label_states had no effect.
    • Each output now lists the target state of every pKa site, in detection order.

Neutralization of charged inputs

  • Charge-separated sulfoxides, sulfones, phosphine oxides, and phosphates (e.g., C[S+](C)[O-]) are collapsed to S=O and P=O instead of protonated to hydroxysulfonium and hydroxyphosphonium cations.
  • Thiolates ([S-]) are protonated, so thiol and thioic acid sites are found.
  • Amidinium, guanidinium, imidazolium, and pyrazolium cations drawn with the charge on a nitrogen without H (e.g., C[n+]1cc[nH]c1) are neutralized.
  • N-oxides and nitrones keep their O- instead of becoming N-hydroxy cations.
  • Nitrate gets one proton, not two, so it is deprotonated again at physiological pH.
  • The terminal N- of a diazo group is left alone, so diazo compounds are no longer permanent +1 cations.
  • A rule that keeps matching its own product raises an error instead of looping forever.

Site detection

  • Two matches of one pattern can no longer claim the same site.
    • A primary amide nitrogen was listed and charged twice, and phosphoric acid lost all three protons.
  • A carbon used by one group as its attachment point no longer hides another group's site.
    • The amine of NCP(=O)(O)O was never protonated.
  • A nitro group no longer hides the site on the atom it is attached to.
    • The N-H of N-nitrourethane (pKa 3.28) was never deprotonated, because the nitro pattern matched that nitrogen as context and protected it.
    • The pattern now requires the substituent without matching it, so both sites are found on the one molecule.

Specific compound classes

  • Nitroguanidines were returned as cations.
    • A nitro group all but removes guanidine basicity (nitroguanidine has a pKaH of -0.98, against 13.6 for guanidine), but the general guanidine entry assigned them 12.03.
    • Nitroarginine and related compounds are affected.
  • N-nitroimines were returned in both states.
    • The neonicotinoid insecticides carry the nitro group on the imine nitrogen, where it costs about twelve units of basicity (imidacloprid's conjugate acid is 1.56), but the general amidine entry assigned them 10.04.
    • Thiamethoxam, imidacloprid, clothianidin, nitenpyram, and dinotefuran are affected.
  • Nucleobases were returned in many protonation states at physiological pH.
    • No entry covered their ring nitrogens, so each fell to the generic aromatic-nitrogen rules, whose windows reach the default pH range: ATP, GMP, and GTP gave 16 variants, and CMP, UMP, and dTMP gave 4, where each should give 2.
    • At low pH the same rules protonated every ring nitrogen and the exocyclic amine together, returning adenine as a tetracation where only N1 is protonated.
    • New entries for 9-substituted adenine and guanine and 1-substituted cytosine and uracil (including thymine) now assign one site per measured pKa:
      • adenine N1 (pKaH 3.99)
      • guanine N7 (pKaH 2.61) and N1-H (pKa 9.45)
      • cytosine N3 (pKaH 4.33)
      • uracil N3-H (pKa 9.60)
    • The remaining ring nitrogens and the exocyclic amines are claimed so that no generic rule reaches them.
    • Nucleosides, nucleotides, and cofactors such as NAD, CoA, and SAM are affected, so pipelines that depend on the previous enumeration should be checked.
    • The values come from published measurements and the IUPAC Digitized pKa Dataset, cited in the substructure file.
    • Free bases (N9-H or N1-H), 5-halo pyrimidines, fused analogs, and hypoxanthine keep the generic rules.

Invalid and labeled structures

  • A protonation state that cannot be kekulized (e.g., a +1 charge on a bridgehead aromatic nitrogen) now keeps the site's previous state.
    • Such variants were dropped as invalid SMILES, and when that was the site's only state the molecule vanished from the output.
    • A molecule whose every variant is rejected now falls back to its input.
  • An aromatic N-H written with an atom map number or isotope label (e.g., [nH:1], [15nH]) could not be deprotonated.
    • The extra hydrogen was removed only when the SMILES contained the literal text [nH-], so a labeled site kept it, failed sanitization, and stayed neutral at every pH.
  • Deuterium and tritium on O, N, or S are treated as ordinary hydrogen, so deuterated acids can be deprotonated.
    • Labels on carbon are kept.

Input and CLI

  • A UTF-8 byte order mark at the start of an input file is ignored instead of corrupting the first SMILES.
  • The CLI refuses an --output_file that is the input file, which it previously truncated before reading.
    • Invalid arguments no longer truncate an existing output file, and the output file is closed properly.
  • A SMILES containing / was read as a file path, and failed with "File not found", whenever the text before the last slash named a directory in the working directory (e.g., CC/C=C(C)C next to a directory named CC).
    • A string is now read as a path only if it exists or ends in a recognized extension.

Don't miss a new dimorphite_dl release

NewReleases is sending notifications on new releases.