inst/extdata/nn_params_provenance_v1.1.1.md

In TmCalculator: Genome-Wide Nucleic Acid Melting Temperature Profiling and Multi-Omics Integration

Nearest-neighbor parameters added in v1.1.1: provenance

Provenance record for the parameter sets added in this release: the two raised in review (Banerjee 2020, Zuber 2022) and the two molecular-crowding sets from the same group (Ghosh 2020, 2023). Both were transcribed from the published tables; the "package key" columns give the mapping onto this package's row-name convention (5'XY3' / 3'WZ5', RNA U written as T).

Units: ΔH in kcal/mol, ΔS in cal/(mol·K) (eu). The package stores (ΔH, ΔS) in the two columns left, right and derives ΔG internally, so only ΔH and ΔS are transcribed here.

1. Banerjee et al. (2020) — RNA/DNA hybrid, 100 mM NaCl

Banerjee D, Tateishi-Karimata H, Ohyama T, Ghosh S, Endoh T, Takahashi S, Sugimoto N. Improved nearest-neighbor parameters for the stability of RNA/DNA hybrids under a physiological condition. Nucleic Acids Research 2020;48(21):12042–12054. doi:10.1093/nar/gkaa572 — Table 2.

Status: implemented as RNA_DNA_NN_Banerjee_2020 (v1.1.1), with attr(, "salt_mM") = 100 so salt correction is skipped automatically at the fitted condition.

Paper keys are written rXY/dWZ with the DNA strand 5'→3'; the package writes the bottom strand 3'→5', so rAC/dGT → AC/TG.

| Paper key | Package key | ΔH | ΔS | |---|---|---|---| | rAA/dTT | AA/TT | −7.8 | −22.9 | | rAC/dGT | AC/TG | −10.1 | −27.3 | | rAG/dCT | AG/TC | −9.4 | −26.2 | | rAU/dAT | AT/TA | −5.8 | −17.5 | | rCA/dTG | CA/GT | −9.8 | −27.4 | | rCC/dGG | CC/GG | −9.5 | −24.8 | | rCG/dCG | CG/GC | −9.0 | −24.3 | | rCU/dAG | CT/GA | −6.1 | −17.9 | | rGA/dTC | GA/CT | −8.6 | −22.7 | | rGC/dGC | GC/CG | −10.6 | −27.7 | | rGG/dCC | GG/CC | −13.3 | −35.7 | | rGU/dAC | GT/CA | −9.3 | −25.5 | | rUA/dTA | TA/AT | −6.6 | −19.7 | | rUC/dGA | TC/AG | −6.5 | −16.3 | | rUG/dCA | TG/AC | −8.9 | −23.3 | | rUU/dAA | TT/AA | −7.4 | −24.3 | | init. rG−dC / rC−dG | init_G/C | 0 | −4.9 | | init. rA−dT / rU−dA | init_A/T | 0 | −7.0 |

Audit correction (2026-09-28). The stored stack values match Table 2, but the initiation mapping does not reproduce the published conditions. Footnote b specifies one initiation for a duplex with at least one GC end; footnote c specifies the alternative when both ends are AT. The package currently applies the listed values at each end. Thus two GC ends contribute −9.8 entropy units instead of −4.9; a mixed pair of ends contributes −11.9 instead of −4.9; two AT ends contribute −14.0 instead of −7.0. Numerical behavior is unchanged in this citation audit; a model correction remains open.

Consistency check. For the sixteen stacks, ΔH − 310.15·ΔS/1000 reproduces the paper's tabulated ΔG°37 to the printed precision (e.g. rGG/dCC: −13.3 + 11.07 = −2.23 ≈ −2.3). The two initiation rows do not: the paper lists ΔG°37 = +2.0 and +2.6, whereas ΔH = 0 with ΔS = −4.9 and −7.0 gives +1.52 and +2.17. Since the package computes Tm from ΔH and ΔS, the tabulated ΔH/ΔS values are used as printed. This discrepancy appears to be in the source table.

2. Zuber et al. (2022) — RNA/RNA, improved end effects

Zuber J, Schroeder SJ, Sun H, Turner DH, Mathews DH. Nearest neighbor rules for RNA helix folding thermodynamics: improved end effects. Nucleic Acids Research 2022;50(9):5251–5262. doi:10.1093/nar/gkac261 — Tables 1A and 1B, "New Model" columns.

Status: implemented as RNA_NN_Zuber_2022 (stacks, Tables 1A and 1B) together with its companion RNA_NN_Zuber_2022_END (penultimate-pair dependent end terms). See "End effects" below for how the end terms are applied, and section 2C for the one term that is deliberately omitted.

Despite the title — the paper says "helix folding", which in this literature means helix formation from two separated strands rather than intramolecular folding — these are duplex parameters: they are fitted from optical melting of short RNA duplexes using the two-state relation 1/Tm = R·ln(Ct/4)/ΔH° + ΔS°/ΔH°, and the comparison column of Table 1A is Xia et al. (1998), i.e. this set is a direct successor to RNA_NN_Xia_1998. The stacks are therefore usable for duplex Tm without modification.

2A. Watson–Crick–Franklin stacks (Table 1A)

The paper's convention (top 5'→3' / bottom 3'→5') is the package's convention, so mapping is U→T plus, for three stacks, the equivalent representative already used by the package (a stack read from the opposite strand is the same stack).

| Paper key | Package key | ΔH | ΔS | |---|---|---|---| | AA/UU | AA/TT | −7.44 | −20.98 | | AU/UA | AT/TA | −8.91 | −25.22 | | UA/AU | TA/AT | −9.16 | −25.40 | | CA/GU | CA/GT | −10.47 | −27.08 | | AC/UG | GT/CA | −11.98 | −31.37 | | AG/UC | CT/GA | −9.34 | −23.66 | | GA/CU | GA/CT | −13.75 | −36.53 | | CG/GC | CG/GC | −9.61 | −23.46 | | GC/CG | GC/CG | −16.52 | −42.13 | | CC/GG | GG/CC | −13.94 | −34.41 | | Initiation | init | +4.66 | +1.78 | | Symmetry | sym | 0 | −1.38 | | AU End on AU | (see note) | +4.36 | +13.35 | | AU End on CG | (see note) | +3.17 | +8.79 |

Verified: ΔH − 310.15·ΔS/1000 reproduces every tabulated ΔG°37 (e.g. GC/CG: −16.52 + 13.07 = −3.45 ≈ −3.46; Initiation: 4.66 − 0.55 = 4.11 ≈ 4.10). The three re-mapped rows are confirmed against the paper's own "1998 Model" column, which matches the values already stored in RNA_NN_Xia_1998 (AC/UG ↔ our GT/CA = −11.40; AG/UC ↔ our CT/GA = −10.48; CC/GG ↔ our GG/CC = −13.39).

End effects

Xia 1998 applies one terminal-AU penalty per AU end, which the package stores in init_A/T (3.72, 10.5) and applies as init_A/T × (number of AT ends). Zuber 2022 replaces this with terms that depend on the penultimate base pair: +4.36/+13.35 when the AU end sits on an AU pair, +3.17/+8.79 when it sits on a CG pair, plus four further values for GU ends. Entering any single value in init_A/T would misrepresent the model, so init_A/T is zero for this set and the end terms are carried by a companion table instead.

The condition "terminal AU pair with penultimate X pair" is exactly a property of the terminal dinucleotide stack, and the core already computes those keys (keys_fr[1] and .right_key()) for the dangling-end and terminal-mismatch lookups. RNA_NN_Zuber_2022_END is looked up on those same keys, at both duplex ends, and its contribution is added without trimming the terminal base pair — unlike de_tbl and tmm_tbl, which consume it. Getting that distinction wrong produces a systematic offset rather than an obvious error, so it is covered by a dedicated test (tests/testthat/test_tm_nn_rcpp.R, "Zuber 2022 end effects are applied at both duplex ends"), and the compiled and reference implementations are checked against each other for this set.

A parameter set supplies end terms simply by defining a table named <set>_END; get_end_table() returns it when present and an empty matrix otherwise, so sets without end terms are unaffected.

2B. GU-containing stacks (Table 1B)

| Paper key | Package key | ΔH | ΔS | |---|---|---|---| | GC/UG | GT/CG | −14.73 | −40.32 | | CU/GG | CT/GG | −9.26 | −23.64 | | GG/CU | GG/CT | −12.41 | −34.23 | | CG/GU | CG/GT | −5.64 | −14.83 | | AU/UG | AT/TG | −9.23 | −27.32 | | GA/UU | TT/AG | −10.58 | −32.19 | | UG/GU | TG/GT | −8.76 | −27.04 | | UA/GU | TG/AT | −2.72 | −8.08 | | GG/UU | GG/TT | −9.06 | −28.57 | | GU/UG | GT/TG | −7.66 | −24.11 | | AG/UU | AG/TT | −5.10 | −16.53 | | AU End on GU | (end table) | +5.16 | +18.96 | | GU End on CG | (end table) | +3.91 | +12.17 | | GU End on AU | (end table) | +3.65 | +12.78 | | GU End on GU | (end table) | +6.23 | +22.47 |

These eleven stacks are the same eleven GU rows carried by RNA_NN_Chen_2012, under the strand-symmetry equivalences noted above (e.g. paper GC/UG ≡ package GT/CG), so Table 1B slots into the existing row set. The four GU end terms are held in RNA_NN_Zuber_2022_END alongside the AU ones.

2C. No special GGUC/CUGG correction in the updated model

Zuber et al. explicitly find that the improved end effects make the special non-nearest-neighbor treatment unnecessary. The parenthetical −32.49/−92.57 row describes an alternative fit including a separate motif term; it is not an obligatory extra contribution to the implemented model. The earlier statement that omitting it necessarily understabilizes this motif was wrong. The older Chen (2012) model does treat this motif separately; see the full reference audit.

3. Where the Zuber 2022 tables live

An earlier revision of this document carried a draft table for pasting into R/zzz.R, in which init_A/T held the "AU end on CG" value as a stand-in because the core could not yet express penultimate-pair dependent ends. That draft has been removed: it is not what shipped, and using it would apply a single averaged penalty at both ends and at GU ends, which is precisely the approximation Zuber 2022 exists to replace.

The authoritative definitions are in R/zzz.R:

• RNA_NN_Zuber_2022 — stacks from Tables 1A and 1B, with init_A/T set to zero because the end terms are not a single averaged penalty.
• RNA_NN_Zuber_2022_END — the 24 end rows (12 AU-end, 12 GU-end), keyed on the terminal dinucleotide with the terminal pair written first, so one table serves both duplex ends.

4. Ghosh et al. (2023) — RNA/RNA under molecular crowding

Ghosh S, Takahashi S, Banerjee D, Ohyama T, Endoh T, Tateishi-Karimata H, Sugimoto N. Nearest-neighbor parameters for the prediction of RNA duplex stability in diverse in vitro and cellular-like crowding conditions. Nucleic Acids Research 2023;51(9):4101-4111. doi:10.1093/nar/gkad020 — Table 1 (40 wt% PEG200, 100 mM NaCl).

Status: implemented as RNA_NN_Ghosh_2023_PEG200, salt_mM = 100.

Paper keys match the package convention directly (U written as T), in the same order as nn_row_std.

| Package key | ΔH | ΔS | |---|---|---| | AA/TT | −10.0 | −30.4 | | AT/TA | −10.1 | −30.8 | | TA/AT | −11.1 | −31.5 | | CA/GT | −12.1 | −32.1 | | GT/CA | −10.7 | −28.7 | | CT/GA | −11.2 | −30.4 | | GA/CT | −11.7 | −30.5 | | CG/GC | −11.1 | −28.8 | | GC/CG | −13.8 | −34.6 | | GG/CC | −14.8 | −38.4 | | init | +4.6 | −2.9 | | init_A/T (per terminal AU) | +6.5 | +18.2 | | sym (self-complementary) | 0 | −1.4 |

All twelve tabulated ΔG°37 values are reproduced by ΔH − 310.15·ΔS/1000 to the printed precision.

5. Ghosh et al. (2020) — DNA/DNA under molecular crowding

Ghosh S, Takahashi S, Ohyama T, Endoh T, Tateishi-Karimata H, Sugimoto N. Nearest-neighbor parameters for predicting DNA duplex stability in diverse molecular crowding conditions. PNAS 2020;117(25):14194-14201. doi:10.1073/pnas.1920886117 — Supporting Information Tables S7 (enthalpy) and S8 (entropy), 40 wt% PEG200 with 100 mM NaCl.

Status: implemented as DNA_NN_Ghosh_2020_PEG200, salt_mM = 100.

Important transcription note. Tables S7 and S8 have two parameter columns. The first, ΔH°/ΔS°[cation], is the no-cosolute reference at 100 mM NaCl (identical to Table S3). The second, ΔH°/ΔS°[40 wt% PEG 200], is the crowding increment, not an absolute value. The values below are the sum of the two columns; the sign of a fitted individual term alone does not establish whether it is an increment. The 2026-09-28 audit verified the article identity but could not reopen the original SI; these S7/S8 transcriptions remain pending independent verification.

| Package key | ΔH[cation] | ΔH increment | ΔH used | ΔS[cation] | ΔS increment | ΔS used | |---|---|---|---|---|---|---| | AA/TT | −7.9 | +1.4 | −6.5 | −23.3 | +4.1 | −19.2 | | AT/TA | −7.2 | −2.2 | −9.4 | −21.3 | −8.1 | −29.4 | | TA/AT | −7.2 | +2.9 | −4.3 | −22.0 | +8.7 | −13.3 | | CA/GT | −8.5 | −4.6 | −13.1 | −23.4 | −15.4 | −38.8 | | GT/CA | −8.4 | −0.8 | −9.2 | −23.2 | −3.6 | −26.8 | | CT/GA | −7.8 | +4.4 | −3.4 | −21.5 | +13.6 | −7.9 | | GA/CT | −8.2 | +3.3 | −4.9 | −23.4 | +10.4 | −13.0 | | CG/GC | −10.6 | +4.2 | −6.4 | −28.2 | +12.1 | −16.1 | | GC/CG | −9.8 | +5.6 | −4.2 | −25.0 | +15.7 | −9.3 | | GG/CC | −8.0 | +4.0 | −4.0 | −20.4 | +11.5 | −8.9 | | init_G/C (per GC end) | +0.1 | −10.2 | −10.1 | −2.8 | −32.3 | −35.1 | | init_A/T (per AT end) | +2.3 | −5.2 | −2.9 | +4.1 | −16.8 | −12.7 | | sym | — | — | 0 | — | — | −1.4 |

Validation: re-predicting duplexes from Table S4 with the derived values reproduces the paper's own predicted column, e.g.

| Duplex | ΔH derived | ΔH published | ΔS derived | ΔS published | |---|---|---|---|---| | d(GGCAGTTC) | −65.5 | −65.7 | −194.1 | −195.0 | | d(CGCTGTAG) | −64.2 | −64.3 | −190.3 | −191.4 | | d(GGACGTCC) (self-comp.) | −62.8 | −63.0 | −185.1 | −185.7 | | d(CGTCGACG) (self-comp.) | −67.6 | −67.9 | −199.5 | −200.6 | | d(ATGAGCTCAT) (self-comp.) | −71.6 | −71.6 | −214.3 | −215.1 |

Residual differences come from rounding the one-decimal tabulated inputs across nine or more summed terms.

Initiation in this model is per duplex end (per G-C end and per A-T end), matching the package's init_G/C / init_A/T handling; the global init term and the oneG/C, allA/T, 5T/A terms are therefore zero.

TmCalculator documentation built on Oct. 5, 2026, 5:08 p.m.