“Determining absolute ligand affinities from fluorine NMR chemical shift anisotropy” sounds like the kind of phrase that makes normal humans suddenly remember an urgent dentist appointment. But buried inside that gloriously niche title is a practical detective story: how do you tell whether a tiny drug fragment is barely shaking hands with a protein, or quietly clinging on like it knows where the snacks are?
The paper in question, CSAKD: Determining Absolute Ligand Affinities From 19F NMR Chemical Shift Anisotropy, reports a new way to measure binding strength in fragment-based drug discovery using fluorine NMR relaxation, chemical shielding anisotropy, and a machine-learning shortcut for predicting shielding tensors. According to the authors, the method can estimate absolute dissociation constants, or KD values, without the usual titration experiments or isotopic protein labeling Rüdisser et al., 2026.
That may sound like lab-instrument inside baseball. It is. But it is also potentially very useful inside baseball.
The Fragment Problem: Tiny Molecules, Big Attitude
Fragment-based drug discovery starts with small molecules. Not full-blown drug candidates, more like chemical LEGO bricks with commitment issues. These fragments usually bind weakly to a protein target, then medicinal chemists grow, merge, or polish them into stronger compounds.
The trouble is that weak binding is hard to measure cleanly. A fragment may bind in the millimolar range, which is scientific code for “yes, technically, if you squint.” Traditional biophysical methods such as NMR, surface plasmon resonance, and calorimetry can help, but follow-up affinity measurements often mean titration series, careful concentration control, and enough pipetting to make your thumb file a workplace complaint.
The headline claim of CSAKD is simple: use information already hiding in fluorine NMR relaxation to infer how tightly a fragment binds.
Follow the Fluorine
Fluorine-19 NMR has become a favorite tool in fragment screening because fluorine is wonderfully nosy. Biological systems usually contain very little fluorine naturally, so a fluorinated fragment gives a clean signal rather than shouting over the entire molecular orchestra. Reviews of 19F NMR in fragment-based discovery have emphasized exactly this point: fluorine can track weak ligand-target interactions with high sensitivity and minimal background noise Li and Kang, 2024.
CSAKD builds on earlier work from the same research line, including CSAR and FastCSAR, methods that used chemical shift anisotropy to rank fluorinated ligands by affinity Rüdisser et al., 2020. The new twist is stronger: not just ranking fragments, but estimating absolute KD values.
KD is the number that tells you how easily the protein-ligand complex falls apart. Lower KD means tighter binding. Higher KD means the ligand is mostly loitering nearby, pretending it has plans.
The Anisotropy Clue
In NMR, a chemical shift is not just a single neat number, even though spectra often make it look that way. The local electron cloud shields a nucleus differently depending on molecular orientation. That directional dependence is called chemical shift anisotropy, or CSA. In liquids, tumbling averages much of this behavior, but CSA still affects relaxation, especially for nuclei like fluorine.
Here is the investigative trail: when a small fluorinated ligand is free in solution, it tumbles quickly. When it binds to a large protein, it tumbles much more slowly. That changes relaxation behavior. CSAKD uses those relaxation effects, together with the ligand’s chemical shielding tensor, to infer how much ligand is bound and therefore estimate KD.
The tensor part is where things normally get expensive. Calculating chemical shielding tensors can require quantum chemistry, which is accurate but not exactly “quick snack before lunch” computation. The authors therefore add a machine-learning model to predict those tensors faster. Machine learning here is not a crystal ball. It is more like a very caffeinated lookup assistant trained to recognize patterns in molecular environments.
The Numbers Ask an Awkward Question
The paper’s appeal is not that it replaces all affinity measurements. It does not. The real question is narrower and sharper: can CSAKD reduce the bottleneck after an NMR fragment screen, when researchers need to decide which weak hits deserve more chemistry?
Recent work shows the field is already hungry for faster scoring. A 2025 JACS paper on ML-boosted SHARPER NMR reported high-throughput KD estimation from reduced titration data, specifically because full NMR affinity follow-up is slow Nepravishta et al., 2025. Meanwhile, deep-learning reviews in protein-ligand affinity prediction keep warning that computational affinity estimates remain hard to generalize, benchmark, and interpret Wang, Wu, and Wang, 2024; Wang et al., 2024.
That context matters. CSAKD is not another docking score wearing a lab coat. It is experimentally anchored. The machine learning helps with a physical parameter, while the NMR data still comes from the molecule in solution. That is a more grounded setup than asking a neural network to guess chemistry from a static pose and vibes.
What Could Go Wrong?
Plenty. The method depends on accurate shielding tensor predictions, clean relaxation measurements, appropriate fluorinated ligands, and assumptions about binding behavior. If a ligand binds multiple sites, aggregates, changes conformation, or generally acts like a molecule with a secret life, the analysis could get messy.
Also, fragment discovery lives in the land of weak signals and subtle artifacts. The numbers may look tidy in a figure, but the lab bench often tells a different story, usually while someone is recalibrating an instrument and questioning their career choices.
Still, if CSAKD proves reproducible across more targets and fragment libraries, it could make NMR-based fragment workflows faster and less labor-intensive. That means chemists could triage weak binders earlier, spend less time on doomed candidates, and move promising fragments toward structural biology and medicinal chemistry with fewer detours.
The Case So Far
CSAKD is intriguing because it turns a nuisance-like NMR detail into a measurement tool. Chemical shift anisotropy, usually the kind of phrase that clears a room, becomes evidence. Fluorine becomes the informant. Machine learning handles part of the paperwork.
The result is not magic. It is a clever combination of physics, spectroscopy, and prediction aimed at one annoying drug discovery question: among these tiny weak binders, which ones are worth chasing?
That is not flashy. It is better than flashy. It is useful, assuming the next wave of validation holds up.
References
-
Simon H. Rüdisser, Gabriela Stadler, and Alvar D. Gossert. “CSAKD: Determining Absolute Ligand Affinities From 19F NMR Chemical Shift Anisotropy.” Angewandte Chemie International Edition, 2026. DOI: 10.1002/anie.6036832. PMID: 42283365
-
Qingxin Li and Congbao Kang. “Perspectives on Applications of 19F-NMR in Fragment-Based Drug Discovery.” Molecules, 2024. DOI: 10.3390/molecules29235748
-
Simon H. Rüdisser et al. “Efficient Affinity Ranking of Fluorinated Ligands by 19F NMR: CSAR and FastCSAR.” Journal of Biomolecular NMR, 2020. DOI: 10.1007/s10858-020-00325-x
-
Ridvan Nepravishta et al. “Fast and Reliable NMR-Based Fragment Scoring for Drug Discovery.” Journal of the American Chemical Society, 2025. DOI: 10.1021/jacs.5c11092
-
Debby D. Wang, Wenhui Wu, and Ran Wang. “Structure-Based, Deep-Learning Models for Protein-Ligand Binding Affinity Prediction.” Journal of Cheminformatics, 2024. DOI: 10.1186/s13321-023-00795-9
-
Huiwen Wang et al. “Prediction of Protein-Ligand Binding Affinity via Deep Learning Models.” Briefings in Bioinformatics, 2024. DOI: 10.1093/bib/bbae081
Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.