NLM DIR Seminar Schedule

Seminars Home
Schedule Seminar

UPCOMING SEMINARS

RECENT SEMINARS


The NLM DIR holds a public weekly seminar series for NLM trainees, staff scientists, and investigators to share details on current and exciting research projects at NLM. Seminars take place on Tuesdays at 11:00 AM, EST and some Thursdays at 3:00 PM, EST. Seminars are held in the B2 Library of Building 38A on the main NIH campus in Bethesda, MD.

To schedule a seminar, click the “Schedule Seminar” button to the right, select an appropriate date on the calendar to sign up, and then complete the form. You will need an NIH PIV card to access the “Schedule Seminar” page.

Please include seminars by invited visiting scientists in the NLM DIR seminar series. These need not be on a Tuesday or Thursday.

If you would like to schedule a seminar by a visiting scientist, click the “Schedule Seminar” and complete the form. Contact NLMDIRSeminarScheduling@mail.nih.gov with questions. Please follow this link to subscribe/unsubscribe to/from the NLM DIR seminar mailing list.

Titles and Abstracts for Upcoming Seminars


(based on the current date)

Devlina Chakravarty
Oct. 20, 2026 at 11 a.m.

Toward Reliable Amyloid Assembly Prediction

Across diverse organisms, evolution has harnessed tightly regulated amyloid formation for important biological functions, and this process is also associated with various human diseases1. Reliable structure prediction could reveal how sequence and environment determine fibril architecture and prioritize assembly formation. Amyloid fibrils remain challenging because their repetitive assemblies are polymorphic: predictions may reproduce local beta-rich structure while missing the full assembly, and comparison with a single experimental structure may misclassify an alternative polymorph as a failure. We benchmarked five predictors—AlphaFold 32 with/without templates, Boltz-23, ColabFold4, and ESMFold25—on 90 experimental targets6, sampling 25 predictions per condition (11,250 models). We searched the benchmark and Amyloid Atlas⁶ for known polymorphs and evaluated partial matches separately. Across methods, 38 targets (42.2%) had at least one full-assembly recovery, 8 (8.9%) had only partial or subassembly recovery, and 44 (48.9%) had neither. Known-polymorph searches added only two full-assembly recoveries, suggesting that most failures reflect an inability to recover known higher-order organization rather than selecting a documented alternative polymorph. Combining local-structure and interface confidence—(i) mean pLDDT (predicted Local Distance Difference Test) and (ii) ipTM (interface predicted TM-score)—enriched for correct assemblies, but some high-confidence predictions still had incorrect topologies or interfaces. Among 88 energy-matched targets, stabilization energy per molecule⁶ was moderately associated with recovery class (Spearman ρ=0.49) but did not distinguish recovered from unrecovered assemblies. These results motivate a reference-free framework integrating geometry, energetics, and calibrated confidence to identify amyloid-compatible predictions.

Reference:
1. Fowler DM et al. Trends Biochem Sci. 2007; 32:217–224.
2. Abramson J et al. Nature. 2024; 630:493–500.
3. Passaro S et al. bioRxiv. 2025.
4. Mirdita M et al. Nat Methods. 2022; 19:679–682.
5. Candido S et al. bioRxiv. 2026.
6. Sawaya MR et al. Cell. 2021; 184:4857–4873.

Bayarbaatar Amgalan
Oct. 27, 2026 at 11 a.m.

TBD

TBD

Joseph Thole
Nov. 3, 2026 at 11 a.m.

TBD

TBD

All Seminars

ARCHIVES